SKIN COLOUR, PIGMENTATION AND THE PERCEIVED HEALTH OF HUMAN FACES. Ian D. Stephen

Transcription

1 SKIN COLOUR, PIGMENTATION AND THE PERCEIVED HEALTH OF HUMAN FACES Ian D. Stephen A Thesis Submitted for the Degree of PhD at the University of St Andrews 2009 Full metadata for this item is available in Research@StAndrews:FullText at: Please use this identifier to cite or link to this item: This item is protected by original copyright This item is licensed under a Creative Commons License

2 Skin Colour, Pigmentation and the Perceived Health of Human Faces Ian D. Stephen This thesis was submitted for the degree of Doctor of Philosophy in December 2008 i

3 I, Ian David Stephen, hereby certify that this thesis, which is approximately 29,000 words in length, has been written by me, that it is the record of work carried out by me and that it has not been submitted in any previous application for a higher degree. Date Signature of Candidate I was admitted as a candidate for the degree of Doctor of Philosophy in January 2006; the higher study for which this is a record was carried out in the University of St Andrews between 2006 and Date Signature of Candidate I hereby certify that the candidate has fulfilled the conditions of the Resolution and Regulations appropriate for the degree of Doctor of Philosophy in the University of St Andrews and that the candidate is qualified to submit this thesis in application for that degree. Date Signature of Supervisor Embargo on both all or part of printed copy and electronic copy for the same fixed period of two (2) years on the following ground: Publication would preclude future publication. Date Signature of Candidate Signature of Supervisor ii

4 Acknowledgements I would like to thank first of all Prof. Dave Perrett for supervision, guidance and support, and without whom this project would not have been possible. I would also like to thank the other members of the Perception Lab for their help and inspiration, through the many stimulating conversations that occur in the laboratory. Especially, Dr Jamie Lawson, Michael Stirrat, Vinet Coetzee, Dr Janek Lobmaier, and Dr Johannes Schindelin. Special thanks to Lesley Ferrier for her organisational support and guidance; Michael Stirrat, Dr Johannes Schindelin and Pete Wilcox for technical support and expertise; Prof Anya Hurlbert and Dr Yazhu Ling for guidance and advice on colour calibration; Vinet Coetzee and Jaco Greeff for cross-cultural assistance; Dr Joanne Cecil for diet assessment advice Thanks also to my friends, family and fellow PhD students in the department, and most importantly to Marissa Parkin for supporting me and waiting (mostly) patiently for all this work to be completed. This work was supported by the BBSRC and Unilever Research. iii

5 Table of Contents Acknowledgements... iii Abstract...1 Section 1: Literature Review Introduction Sexual Selection Parental investment and who chooses Health Signalling Handicap Hypothesis Heterozygosity Sexual selection in humans Facial Attractiveness Symmetry Averageness Sexual Dimorphism Masculinity as a handicapping signal Skin Colour Structure and function of human skin Skin Pigmentation Blood Carotenoids Melanin The evolution of Primate Trichromatic Colour Vision Summary...36 Section 2: Carotenoids Affect Skin Colour Section 2 Introduction Study 1 Natural dietary carotenoid intake contributes to skin yellowness Introduction : Predictions Methods : Statistical Methods Results Discussion Study 2 Reflectance Spectra Introduction Methods Statistical Methods and Results Discussion Section 2 Discussion...57 Section 3: Pure Colour Transforms Section 3 Introduction Lightness Redness Yellowness Contrast between skin and lips Methods Photography Colour Transformations Experimentation...73 iv

6 3.2.4 Statistical Methods Results Relative Lip Colour Lightness Redness Yellowness Section 3 Discussion Contrast between Skin and Lips Lightness Redness Yellowness Gradients...82 Section 4: Blood Colour Transforms Introduction Blood Perfusion Blood Oxygenation Primate Colour Signals Methods Photography Empirical Measurement of Deoxygenated Skin Blood Colour Empirical Measurement of Oxygenated Skin Blood Colour Image Manipulation Experimentation Statistical Methods Results Single-Axis Transforms Two-dimensional Colour Transform Section 4 Discussion...97 Section 5: Melanin and Carotenoid Transforms Carotenoid and melanin provide perceptible cues to health in human faces Introduction Carotenoids Melanin Predictions Methods Photography Empirical Measurement of Carotenoid Colour Empirical Measurement of Melanin Colour Image Manipulation Experimentation Statistical Methods Results Single-Axis Transforms Two-Dimensional Transform Discussion Carotenoids Melanin Two-Dimensional Transforms Conclusion Section 6: Cross-Cultural Studies v

7 6.1 Introduction Methods Photography Image Manipulation Experimentation Statistical Methods Results Discussion Section 7: General Discussion Effects on Evolution Conclusions References Appendix A: Graphs showing relationship between correlation coefficients and carotenoid absorption spectra Appendix B: Display Accuracy vi

8 Abstract Many non-human animal species use colour to signal dominance, condition or reproductive status. These signals have not previously been noted in humans. This thesis investigates the effects of skin colouration and pigmentation on the apparent health of human faces. Section 2 showed that individuals with increased fruit and vegetable and carotenoid consumption have yellower skin (Study 1) due to increased carotenoid pigmentation in the skin (Study 2). In Section 3, participants enhanced the redness, yellowness and lightness of the skin portions of colour-calibrated facial photographs to optimise healthy appearance. This suggests roles for blood (red) and carotenoid/melanin (yellow) colouration in providing perceptible cues to health. The contrast between lips and facial skin colour was not found to affect the apparent health of the faces, except in the b* (yellowness) axis, where enhanced facial yellowness caused an apparent blue tint to the lips. In Section 4 participants enhanced empirically-derived oxygenated blood colour more than deoxygenated blood colour to optimise healthy appearance. In two-dimensional trials, when both blood colour axes could be manipulated simultaneously, deoxygenated blood colour was removed and replaced with oxygenated blood colour. Oxygenated blood colouration appears to drive the preference for redness in faces. 1

9 In Section 5 participants increased carotenoid colour significantly more than they increased melanin colour in both single-axis and two-dimensional trials. Carotenoid colour appears to drive the preference for yellowness in faces. In a cross-cultural study (Section 6), preferences for red and yellow in faces were unaffected by face or participant ethnicity, while African participants lightened faces more than UK participants. A preference for more redness in East Asian faces was explained by this group s lower initial redness. The thesis concludes that pigments that provide sexually-selected signals of quality in many non-human animal species carotenoids and oxygenated blood - also provide perceptible cues to health in human faces. 2

10 Section 1: Literature Review 3

11 1.1 Introduction This thesis investigates the role of skin colour and pigmentation in the perception of health in human faces. It is known that colour is used to signal condition (Martinez-Padilla et al., 2007; Olson & Owens, 1998), sexual status (Setchell et al., 2006), dominance (Setchell & Wickings, 2005) and hormonal status (Rhodes et al., 1997) in many non-human animal species. There is evidence that colourful ornaments are condition dependent, with better fed (Perez-Rodrigues & Vinuela, 2008) and less parasitized (Martinez- Padilla et al., 2007; Olson & Owens, 1998) individuals exhibiting larger and brighter colourful ornaments. Section 2 of this thesis examines the role of diet in determining the colour of the skin, particularly in relation to carotenoid pigments (which provide the colour in many of the animal kingdom s colourful ornaments). The remainder of this thesis will address the effect of skin colour and pigmentation on the apparent health of human faces. Colourful signals are often sexually selected, with bigger and brighter ornaments being preferred by the opposite sex (while it is more common for males to exhibit these ornaments, and females to be the choosy sex, examples of female ornamentation and male choice exist; Massaro et al., 2003; Waitt et al., 2003). Many of the colourful displays made by non-human animals involve the bare skin (Perez-Rodrigues & Vinuela, 2008). Skin colour in humans is one of the most widely varying physical characteristics, both between and also within populations (Jablonski, 2006; Robins, 1991). Skin colour can communicate a number of social and biological messages. People associate more negative stereotypes such as aggression and criminality with 4

12 black people with darker skin than they associate with black people with lighter skin (Maddox & Gray, 2002). Blushing is a reddening of the face that is associated with social embarrassment (Leary, et al., 1992) and also with sexual attraction (Bogels et al., 1996). However, while the distribution of pigmentation colour in human faces has been shown to affect perceptions of age, health and attractiveness (Matts et al., 2007), the overall skin colour has not previously been studied with respect to perceptions of health. I will examine the role of skin colour and pigmentation in providing a perceptible cue to human health, and its possible role in sexual selection. In this Section, I review the relevant literature relating to sexual selection, face perception and skin colour. 1.2 Sexual Selection Many species exhibit phenotypic and behavioural traits that appear to be deleterious in terms of survival. The peacock s tail is large, brightly coloured and cumbersome, simultaneously making the bearer more conspicuous to predators and less capable of evading predation. Why then should such phenotypes evolve? Darwin (1871) proposed a theory of sexual selection to explain them. Perhaps individuals with larger and brighter ornaments are preferred as mates by the opposite sex. Under this hypothesis, the increased reproductive success from sexual selection outweighs the handicap to survival caused by having such ornaments. In this way, larger and brighter ornaments may be selected for in the population. 5

13 Many animals exhibit traits that appear maladaptive to survival, such as the bright plumage of birds, which may attract predators. These ornaments can be maintained through sexual selection if the bearer has increased reproductive success (Darwin, 1871; Trivers, 1972). Since males have greater variation in their reproductive success than females in most species (Bateman, 1948), access to females is a limiting resource. Females invest more in each offspring than males, right from the initial difference in gamete size (anisogamy) to mammalian gestation and, in many species, post-partum care (Trivers, 1972). Males should therefore be expected to compete for access to females (male-male competition), and females are expected to be choosy, in order to obtain the best mates (female choice; Trivers, 1972). Several studies have confirmed that males who are best able to compete with other males achieve preferential access to females (e.g. the best fighters in elephant seals achieve almost all of the copulations in a colony; Le Boeuf, 1974). Also, males who are best able to attract females by being, for example, the house finch with the brightest red head and throat plumage (McGraw et al., 2001) or the sage grouse with the most extravagant displays (Gibson, 1996), obtain greater numbers of matings. Ornamentation may be preferred by females for a number of reasons. Fisher (1958) proposed that once a male characteristic is preferred, for any reason, the very fact that it is preferred provides an impetus for the continuation and accentuation of that male trait and the corresponding heritable (genetically or culturally; Laland, 1994) female preference for the trait. If bright-coloured males are preferred, a selection pressure exists for males to evolve ever brighter plumage. Females who prefer bright plumage will also obtain a reproductive advantage for their male offspring who will, in turn, obtain more reproductive success. This sexy sons hypothesis is known as Fisherian 6

14 runaway selection (Fisher, 1958). This hypothesis relies on the heritability of preference as well as the heritability of trait 1.3 Parental investment and who chooses Sexual selection is thought to originate from the difference in the parental investment made by the two sexes (Trivers, 1972) and is a phenomenon that is likely to have existed since very early in the evolution of life. The emergence of anisogamy meant that the sex producing the larger gamete (the female) has made a larger initial investment in the embryo than the sex producing the smaller gamete (the male). Indeed, a male can impregnate a female and walk away with very little potential loss of investment. However, in many mammalian species, the female is committed to a substantial investment for at least the length of the gestation period. At birth, the mother must either continue to invest in the offspring or kill or abandon it, which would represent a significant loss of investment (Trivers, 1972). This greater investment by females means that females are expected to be choosier about both mate quality, so as not to waste investment on weak offspring, and also to be choosier about the male s potential and willingness to invest in the offspring. This difference in initial investment means that the resources limiting the reproductive output of the two sexes are different (Bateman, 1948). In Bateman s (1948) original experiments with Drosophila melanogaster, females were found to be limited by the number of eggs they could produce. Males, with much cheaper gametes, were limited by their access to females. This finding has been replicated in a number of other species (see Trivers, 1972). 7

15 In many species, however, investment does not end at birth, and includes any investment by the parent in an individual offspring that increases the offspring s chances of surviving (and hence reproductive success) at the cost of the parent s ability to invest in other offspring, such as incubation, protection, feeding, and teaching. Investment in securing a mate is not included (Trivers, 1972). Many species, including humans, show significant male parental investment. As male investment increases, so male choosiness will increase (at least for females with whom he will invest and not simply desert) as having high quality offspring becomes more important. In species such as humans, where males invest more than half but less than equal to the female investment, selection favours a mixed strategy for both sexes (Parker & Simmons, 1996). Females should be choosy and select a male who has good genes, high investment potential and low likelihood of desertion, but may also increase their fitness through extra-pair copulations (EPCs) with males who are genetically superior to their partner. Similarly, males are expected to invest in the offspring of a single female, whilst competing to inseminate females in whose offspring he will not invest, particularly if those offspring will be raised by another male. Males are expected to be choosy about the female in whose offspring he will invest (Parker & Simmons, 1996; Trivers, 1972). The risk of sexually transmitted pathogens and, in some species, social risks associated with EPCs will induce choosiness in extra-pair partners. In humans, therefore, both males and females are expected to be choosy, and both males and females are expected to compete for mates. 1.4 Health Signalling Attraction is a way to ensure that individuals reproduce with the most beneficial mate and to raise the most successful offspring. Direct benefit models of sexual selection 8

16 suggest that individuals choose mates that offer the most benefit directly to themselves. This may involve nuptial gifts, territory and parental care (Trivers, 1972). Indirect benefit models of sexual selection suggest that individuals choose mates based on genetic quality, gaining advantages associated with good genes in their offspring (Fisher, 1958). Choosing mates based upon their health potentially provides both direct and indirect benefits. A healthy mate will be better equipped to provide more parental care and is also likely to have a lower parasite load, exposing the individual to a lower chance of infection (direct benefits). Additionally, any heritable aspect of increased health, such as a better immune system or more efficient heart and lungs, may be passed on to offspring (indirect benefits). The identification of health, and signalling of health, should therefore play a role in mate selection. 1.5 Handicap Hypothesis Another hypothesis for the evolution of seemingly maladaptive traits is known as the handicap hypothesis. Traits which are preferred by one sex advertise quality by imposing a handicap on the bearer (Zahavi, 1975). By investing in a costly ornament, the bearer is advertising that he has the resources to waste, and more. This is a restatement of the economic concept of conspicuous consumption (Getty, 2002; Veblen, 1899). The development of male secondary sexual characteristics requires significant investment in terms of resources (Zahavi, 1975), including energy and useful compounds such as carotenoids (carotenoid-based ornamentation requires the deposition of carotenoids in the skin, hair or feathers, making it unavailable to other processes such as immune function; Saks et al., 2003). The development and maintenance of these costly ornaments can only be achieved by high-quality 9

17 individuals. Females that choose males with large ornaments can be assured that they are obtaining high quality mates (Zahavi, 1975). It was proposed that the quality being advertised by sexual ornaments is freedom from parasites and pathogens (Hamilton & Zuk, 1982). Those males who can better resist parasites and pathogens, or through behaviour encounter fewer parasites and pathogens, can afford to invest more in costly ornamentation to advertise their quality (Hamilton & Zuk, 1982). American goldfinches artificially infected with intestinal coccidians had less saturated carotenoid-based bill and plumage colouration than individuals who were free from infection (McGraw & Hill, 2000), supporting this hypothesis. Folstad and Karter (1992) proposed testosterone as a mediating pathway for the relationship between parasitism and ornamentation. The immunocompetence handicap hypothesis (ICH) proposes that testosterone - required for the development of secondary sexual characteristics - is an immunohandicap, and is affected by nutritional and disease status. It may therefore provide a mechanism for the trade-off between investment in secondary sexual characteristics and reproductive effort on the one hand and immunocompetence on the other (Folstad & Karter, 1992). Only highquality males, or those who pay a lower cost for higher testosterone levels, will therefore be able to produce and maintain large secondary sexual characteristics without compromising their immune system excessively. Ornamentation could then be an indicator of good genes (Folstad & Karter, 1992). 10

18 By mating preferentially with males that display indicators of immunocompetence and freedom from parasites, females obtain direct benefits such as protection from other males (e.g. Utami et al., 2002) and increased resources, or improved breeding grounds (e.g. Parker, 1983) and avoidance of parasites and infectious diseases (Hamilton & Zuk, 1982). They may also obtain indirect benefits in the form of good genes, improved parasite resistance and increased reproductive potential that will only be expressed in the offspring (Fisher, 1958; Grammer et al., 2003). Hamilton and Zuk (1982) suggested that their hypothesis would be supported if individuals with larger ornaments had lower parasite loads. This also extends to individuals with higher testosterone having lower parasite loads (Folstad & Karter, 1992). Studies attempting to find these patterns found mixed results, with some studies finding negative associations between ornament size and parasite load (Moller et al., 1999) and some finding positive associations (Gonzalez et al., 1999; Hamilton & Zuk, 1982). Both have been interpreted as supporting the handicap hypothesis. However, Getty (2002) showed that the ICH does not make predictions in either direction, since the resources of the individual and the magnitude of the handicap compared to these resources will determine the penalty paid for the ornament. Even though larger ornaments result in larger handicaps, high-quality individuals may begin with greater resources and pay a lower relative cost of ornamentation. The relative effects of ornament size, initial resources and relative cost all determine parasite load. Therefore, individuals with large ornaments may have high or low parasite loads (Getty, 2002). 11

19 Two testable predictions are made by the ICH. Increases in testosterone level should inhibit immune function and increase reproductive effort (Folstad & Karter, 1992), and higher quality males should pay a lower immunocompetence price for increased testosterone (Grafen, 1990). Several studies have investigated these predictions, using testosterone implants, and again found mixed results. High testosterone male redwinged blackbirds and males with testosterone implants did not exhibit reduced antibody production in response to inoculation with keyhole limpet haemocyanin in a free-ranging population (Hasselquist et al., 1999). However, testosterone-implanted sand lizards exhibited greater mobility and higher mating success. They also exhibited greater mass loss during the mating season and increased tick load, supporting the ICH (Olsson et al., 2000). In barn swallows, testosterone-implanted males suffered increased ectoparasite load, though this increase was smaller in individuals with longer tails, supporting the hypothesis that testosterone has an immunosuppressive effect and that higher-quality males pay a reduced cost of testosterone (Saino et al., 1995). Testosterone-implanted male red grouse suffered increased parasite load and reduced condition, but also showed increased comb size (Mougeot et al., 2004). Lindstrom et al. (2001) found that testosterone-implanted greenfinches exhibited increased immune response to the common Sindbis virus early in infection, but decreased response later on and overall. There was also weak evidence that males with larger tail patches paid a reduced cost of testosterone, providing mixed results (Lindstrom et al., 2001). These findings provide some support for the hypothesis that high-quality individuals pay a lower immunocompetence cost of bearing large ornaments. 12

20 1.6 Heterozygosity One potential explanation of the good genes that are suggested to be sought in mate choice is heterozygosity (Brown, 1997). Heterozygosity is beneficial in a number of species, as a greater degree of heterozygosity reduces the number of deleterious recessive alleles that are likely to be expressed. Heterozygous salmonid fish exhibit increased growth rate, disease resistance and developmental stability (Brown, 1997; Leary et al., 1985; Mulvey et al., 1994). Red deer show a correlation between heterozygosity at a number of microsatelites and birth weight (Slate & Pemberton, 2002). Females should be expected to prefer dissimilar mates (who may be expected to have different alleles to oneself at a larger number of loci), in order to increase heterozygosity in their offspring. Additionally, particularly in species with high paternal investment, preferences are expected for males with high levels of heterozygosity, as more heterozygous mates are more likely to be healthy, expose the female to lower levels of pathogens and be better able to provide paternal investment in offspring. These preferences have been found in several species (Brown, 1997). The major histocompatibility complex (MHC, in humans known as the human leukocyte antigen, HLA) is a highly polymorphic region of the genome that is involved with parasite and pathogen resistance and self-non-self recognition. It has therefore been studied as a possible location for the action of sexual selection, based on inbreeding avoidance and heterozygote advantage (Brown, 1999; Thornhill et al., 2003; Thornhill & Grammer, 1999). MHC heterozygosity is associated with developmental stability (low fluctuating asymmetry, FA) in many animals and plants, and individuals with low FA are preferred as mates in many species (Brown, 1999), including humans (e.g. Brown et al., 2008; Gangestad et al., 2005; Gangestad et al., 13

21 1994; Grammer & Thornhill, 1994; Mealey et al., 1999; Perrett et al., 1999). Additionally, it has been found that women prefer the scent of men who are heterozygous at MHC alleles (Thornhill et al., 2003) and men who are heterozygous at more MHC loci are rated as more facially attractive by women, perceived to be healthier and perceived to have healthier skin (as rated on patches) than men who are homozygous at certain MHC loci (Roberts et al.., 2005). This suggests that women prefer heterozygosity, perhaps as a mechanism for avoiding catching diseases from parasitized mates, as a way of ensuring better paternal care from healthier partners or for obtaining rare alleles for offspring. Further, there is evidence that people prefer partners who are genetically dissimilar to themselves, which may indicate that they are seeking heterozygosity in their offspring. Hutterites were found to have significantly fewer marriages matching for certain HLA loci than expected by chance (Ober, 1999), and several studies have shown that people prefer the scent of opposite sex individuals with dissimilar MHC to their own (Penn, 2002; Penn & Potts, 1999; Thornhill et al., 2003; Wedekind et al., 1995), though other studies have failed to find these patterns (Hedrick & Black, 1997; Ihara et al., 2000). One prediction of the MHC hypothesis that does not appear to have been addressed relates to the fact that MHC heterozygosity is thought to be an aspect of parasite resistance and good genes (Brown, 1999). If this is the case, MHC heterozygous men should be able to have increased testosterone and increased masculinity, whilst paying a lower fitness cost, though this has yet to be demonstrated. MHC heterozygosity could be a source of quality that the handicap hypothesis can operate on. Since symmetry is thought to be reduced by problems such as illness during development (though direct evidence of this is yet to emerge; Rhodes et al., 2001b), it may be that 14

22 symmetry is a check on the honesty of signals of quality. For example, individuals developing large ornaments without the quality that they are advertising may exhibit greater fluctuating asymmetry. This would explain the greater attractiveness of symmetrical individuals. 1.7 Sexual selection in humans Several studies have shown that humans exhibit many of the behaviours associated with sexual selection, including intrasexual competition and intersexual choice (Symons, 1979). Male-male violence over access to women is widespread. High levels of intra-group male-male violence to obtain and defend access to women has been reported in Aboriginal Eskimos (Rasmussen, 1931). The Yanomamo tribe from Venezuela fight wars with other villages, in order to obtain and defend access to women (Chagnon, 1968). The Mae Enga tribe of Papua New Guinea fight wars over land in order to raise crops and livestock, but the tribesmen report that the reason for obtaining the land is in order to provide food to obtain more wives and raise more children (Meggitt, 1977). Intrasexual competition may also take place in non-violent ways. Politics may provide a way for men to raise their social status and increase access to women (Symons, 1979). Hunting also provides a mechanism for men to compete over women. Despite the relatively small contribution that hunting makes to the nutritional intake of most small societies, men invest significant time and effort into obtaining meat time that, nutritionally, would be better spent foraging for other foods (Symons, 1979). However, meat is exchanged for sex in many societies, with the most successful hunters obtaining the most sexual access to women (Siskind, 1973; Gurven, 2004). 15

23 Indeed, food is implicitly or explicitly required to obtain access to sex in many cultures (Knight, 1991). This suggests that men are using hunting to compete over mates. Intersexual choice also operates in humans, though men and women have been shown to desire different qualities in the opposite sex. Men are consistently more interested than women in the physical attractiveness of potential partners, whereas women are consistently more interested than men in social status and ability and willingness to invest (Buss, 1989). Women become more interested in physical attractiveness when seeking extra-pair copulations (EPCs) and in the fertile phase of the menstrual cycle (Penton-Voak & Perrett, 2000; Penton-Voak et al., 1999). This may be a mechanism to allow women to increase the likelihood of becoming pregnant from an EPC, maximising the good genes benefits, whilst minimising the risk of being caught and punished. The increased preference for masculine men when seeking short term relationships, and preference for feminine men for long term relationships may allow women to maximise the good genes benefits from masculine men, whilst obtaining better parenting and reduced risk of violence from feminine men long term (Little et al., 2002). Men have considerably more variability in their reproductive success than women (a survey of the Yanomamo tribe found that the most reproductively successful man had 43 children, while the most successful woman had 14; Chagnon, 1968). However, the relatively high parental investment made by men in at least some of their offspring means that men are also expected to exercise intersexual choice, especially in women with whom he will invest (Trivers, 1972). Physical attractiveness is a major factor in 16

24 the choice of mates by men (Buss, 1989). While rates of female-female violence are low compared with men, high quality men are a limited resource over which competition takes place, including competition through maximising ones own attractiveness (Buss & Dedden, 1990) and competitor derogation reducing the perceived attractiveness of competitors (Fisher, 2004). Arranged marriages are common in many societies, seemingly removing intersexual choice, especially from women (Crook, 1972). However, intersexual choice is likely to operate in these societies through various mechanisms. EPCs are outside of the control of the elders, and would be subject to the same sexual choice mechanisms that operate in other societies. In addition, younger and more attractive women command higher dowry payments in Kenyan Kipsigis (Borgerhoff Mulder, 1988) and more expensive engagement rings in Americans (Cronk & Dunham, 2007). Families of men therefore prefer to arrange marriages with attractive, young women. Families of women will prefer to arrange marriages with richer men. Pressure to obtain an attractive and wealthy spouse is also likely to be applied by the individuals on their parents or the elders arranging the marriage. 1.8 Facial Attractiveness An important mechanism for mate choice in humans is attractiveness. Whilst the phrase beauty is in the eye of the beholder suggests that social conditioning and individual preferences control attractiveness, a view held by some feminist writers (Wolf, 1991), a growing body of literature suggests that people generally agree on which faces are attractive and which are unattractive, including cross-culturally and in societies with little access to Western media (Cunningham et al., 1995; Langlois et al., 17

25 2000). Additionally, research has identified several specific aspects of faces and bodies that contribute to attractiveness and the aspects of quality that they may signal. Studies on the question of whether aspects of facial attractiveness signals health have yielded mixed results Symmetry Fluctuating asymmetry (the deviance from symmetry in an individual in features that are symmetrical on average across the population) is thought to reflect developmental instability (Van Valen, 1962). Developmental instability is a measure of the ability of an individual to develop stably in the given environmental conditions, with less favourable environmental and genetic factors increasing instability (see Moller & Thornhill, 1998). Indeed, ratings of symmetry have been found to correlate with ratings of attractiveness. Controlling for rated health eliminated most of these effects however, suggesting that the relationship between symmetry and attractiveness is mediated by the healthy appearance of symmetry (Jones et al., 2001; Rhodes et al., 2007). Men with low FA were found to have more sperm per ejaculate, higher sperm speed and better sperm migration (Manning et al., 1998). However, MHC heterozygosity and MHC allelic rarity (thought to be indicators of genetic and immune quality) have been found not to correlate with symmetry (Thornhill et al., 2003). Symmetry (low fluctuating asymmetry) has been found to correlate with attractiveness in several studies in non-human animals (Swaddle & Cuthill, 1994) as well as in human faces (Gangestad et al., 1994; Grammer & Thornhill, 1994; Mealey et al., 1999; Penton-Voak et al., 2001; Perrett et al., 1999) in several cultures (Rhodes 18

26 et al., 2001a) and bodies (Brown et al., 2008; Møller et al., 1995; Singh, 1995). More symmetrical men in a natural fertility population were found to live longer, have more offspring, begin reproduction younger and have more lifetime sexual partners (Waynforth, 1998). Women with more symmetrical partners experience more copulatory orgasms (Thornhill et al., 1995). Since female orgasm is not required for conception, several hypotheses have been advanced to explain its existence (Thornhill et al., 1995). Female orgasm is thought to be a mechanism for moving sperm into the uterus and increasing conception chances (Fox et al., 1970). By having orgasms preferentially with certain men, women can exert some additional influence over the paternity of her offspring (Smith, 1984). The increased frequency of copulatory orgasm with men of low fluctuating asymmetry suggests that symmetry signals good genes (Thornhill et al., 1995). Attraction to other men during the fertile phase of the menstrual cycle is thought to be an adaptation allowing women to maximise genetic fitness benefits by mating with high quality partners during ovulation, whilst minimising the risks of infidelity, and maintaining a long-term relationship. Women with partners with high fluctuating asymmetry experience more attraction to other men during the fertile phase of the menstrual cycle (Gangestad et al., 2005). This suggests that the benefits obtained from extra-pair mating may be higher to those women with more asymmetrical partners. This provides additional evidence that symmetry may signal good genes. Symmetry shows cyclical variation, with women becoming more symmetrical at the follicular (fertile) phase of the menstrual cycle (Manning et al., 1996; though this finding was not replicated in one study of breast volume; Hussain et al., 1999). Men 19

27 who are sensitive to this variation and are more attracted to women at the fertile phase may have more reproductive success. This mechanism also allows women to become more desirable and receive more sexual attention when at their most fertile (Manning et al., 1996). Facial photographs of women taken during the fertile phase of the menstrual cycle are rated as more attractive than those taken in the luteal phase (Roberts et al., 2004). While initial studies found that women were better able to detect facial symmetry in men during menses (women were better able to identify the more symmetrical of two faces during menses than during the luteal phase of the menstrual cycle, suggesting that high progesterone levels may inhibit facial symmetry detection), and failed to find any cyclical change in preference for male facial symmetry (Oinonen & Mazmanian, 2007), another study found that women showed a stronger preference for male facial symmetry during the fertile phase of the menstrual cycle, but only if they already had a partner or were looking for a short-term relationship. This suggests that women are seeking to maximise good genes benefits of short term and extra pair copulations (Little et al., 2007). While a study of medical records failed to link facial symmetry with actual medical health (Rhodes et al., 2001b), a study based on reported number of illnesses found that more symmetrical people had fewer respiratory infections in the preceding three years (Thornhill & Gangestad, 2006b). There have been suggestions that symmetry itself may not be attractive and that it may simply be correlated with another attractive trait. Women prefer the odour of men who are more symmetrical and attractive (Rikowski & Grammer, 1999). Female Japanese scorpionflies show a preference for the pheremones of more symmetrical 20

28 males. When the males were manipulated to be more or less symmetrical, the females chose to copulate with males that were originally more symmetrical regardless of their current symmetry, suggesting that pheremones and not symmetry are controlling mate preference (Thornhill, 1992). A study in humans found that symmetrical faces were more attractive. However, showing only half the face, thereby hiding many cues to symmetry, yielded very similar results, suggesting that symmetry merely correlated with another attractive trait (suggested to be masculinity, Scheib et al., 1999, though this is unclear, Penton-Voak et al., 2001). A further study suggested that asymmetry may in fact be more attractive, perhaps because of an unnatural or unexpressive appearance of symmetrical faces (Swaddle & Cuthill, 1995), though doubt has been cast upon the validity of the methods used in this study, as the mirrored blending method used may have smoothed imperfections in the skin or mirrored blemishes that would not be perfectly symmetrical in real faces (Perrett et al., 1999) Averageness Another aspect of facial attractiveness is averageness. The most common form of natural selection is stabilising selection, in which selection operates against extremes and maintains an equilibrium about the mean (Dobzhansky, 1970; Langlois & Roggman, 1990). Individuals who exhibit characteristics about the population mean are likely to have fewer deleterious mutations and a higher level of heterozygosity than individuals with extreme characteristics (Symons, 1979). Indeed, individuals with more average (less distinctive) faces have been found to have had better childhood health (Rhodes et al., 2001b), though perceived health does not account for all of the attractiveness of average faces (Rhodes et al., 2007). 21

29 Additionally, people respond to prototypical (average) examples of objects as though they were familiar (Strauss, 1979). Familiarity, in turn, increases attraction to faces (Peskin & Newell, 2004). Average objects other than faces (dogs, wristwatches, birds) have also been found to be seen as more attractive, suggesting an inherent attraction to prototypical objects (Halberstadt & Rhodes, 2000). Average composite images are rated as more attractive than most individual images, and the more individual images contribute to the composite image, the more attractive the composite is considered (Langlois & Roggman, 1990), though this finding has not always replicated successfully (Grammer & Thornhill, 1994). However, it has subsequently been suggested that other aspects of the composite images may have contributed to their attractiveness. The process of averaging faces used by Langlois & Rogmann (1990) causes a smoothing of blemishes and an apparent improvement in the skin condition of the faces. However, combining 32 photographs of the same face produced less attractive composites than 32 different faces, suggesting that this cannot explain all of the attractiveness of averageness (Langlois et al., 1994). Composite images are also more symmetrical than individual faces, though controlling for symmetry fails to remove the effect of averageness on attractiveness (Langlois et al., 1994; Rhodes et al., 1999). The effect of averageness on attractiveness has been observed cross-culturally, with both Chinese and Japanese participants preferring faces that had been transformed towards the population mean. Indeed, it made little difference whether the population mean used was of the same or other race (Rhodes et al., 2001a). 22

30 While average faces are attractive, the most attractive faces are not average. Composite faces composed of individual female faces that were rated as highly attractive are found to be more attractive than average composites. The face produced by exaggerating the difference between the average composite and attractive composite to create a face with exaggerated attractive features is found to be more attractive than both the average and attractive composites (Perrett et al., 1994). These attractive composites have more feminine features and may indicate a higher oestrogen/androgen ratio (Johnston & Franklin, 1993). It has been claimed that the attractive composites may be more average than the normal composites. The attractiveness transform may thus be a transform towards more averageness, and this may explain the preference for the attractive transformed faces (Rubenstein et al., 2001). However, another study manipulated faces along an attractiveness dimension, and found that the average composite was rated as the most normal face, while the most attractive face was one that was manipulated towards the attractive composite (though not the most caricatured attractive face; DeBruine et al., 2007) Sexual Dimorphism The degree of sexual dimorphism in faces contributes to their attractiveness. Feminine features of the body develop under the influence of female reproductive hormones, such as oestrogen (Barber, 1995; Singh, 1993), and ratings of femininity correlate with oestrogen levels, suggesting that femininity is a signal of reproductive health (Law Smith et al., 2006). Studies have found that female faces with a high degree of femininity are considered more attractive (Perrett et al., 1998; Rhodes et al., 2003; Rhodes et al., 2007). Perceived health also contributes to the attractiveness of 23

31 feminine female faces (Rhodes et al., 2007) and feminine-faced women experience fewer and less severe respiratory illnesses (Thornhill & Gangestad, 2006a). There have been suggestions that femininity may be a handicapping signal, with oestrogen influencing the development of feminine features and having an immunosuppressant effect (Service, 1998; Singh, 1993). However, since higher oestrogen levels are associated with increased fecundity (Stewart et al., 1993), it is more parsimonious to suppose that feminine features are attractive due to an association with fecundity. Male facial masculinity has a more complicated relationship with attractiveness. The immunocompetence handicap hypothesis suggests that testosterone is an immunosuppressant. Only males with strong immune systems should therefore be able to sustain a high level of testosterone and develop masculine traits (Kirkpatrick & Ryan, 1991). Facial masculinity has been found to be associated with lower levels of respiratory illness and antibiotic use in men (Thornhill & Gangestad, 2006a) and also to appear healthier (Rhodes et al., 2003), supporting the hypothesis and suggesting that masculinity is a signal of immunocompetence that should be considered attractive. Increased testosterone levels are also associated with troubled relationships including increased levels of infidelity, domestic violence and divorce (Booth & Dabbs, 1993), and more masculine faces are perceived as less warm, emotional, honest, cooperative and less good fathers (Perrett et al., 1998). While initial studies failed to find masculine male faces to appear more attractive (Rhodes et al., 2003) or found that women considered more feminine male faces more attractive (Perrett et al., 1998), another study suggests that women s preference for masculinity 24

32 changes across the menstrual cycle, with masculinised faces preferred in the follicular (fertile) phase and feminised faces preferred in the luteal phase (Penton-Voak & Perrett, 2000). Since women s interest in extra-pair copulations also peaks in the fertile phase (Gangestad et al., 2005), this suggests that women may be attracted to faces that signal good genes and immunocompetence when conception is most likely (Penton-Voak & Perrett, 2000), but to faces that indicate more prosocial and parental abilities when conception is less likely. This hypothesis is lent further support by the finding that women prefer more feminine men for long term relationships and more masculine men for short term relationships, perhaps in order to maximise parental investment from feminine men, whilst gaining good genes and immunocompetence benefits from masculine men (Little et al., 2007). Further, women who consider themselves to be more attractive (Little et al., 2001), or who have lower waist-to-hip ratio (a more attractive body shape) or who are rated by others as more attractive prefer more masculine male faces for a long term relationship than their less attractive peers (Penton-Voak et al., 2003). This result is driven by the less attractive women, who prefer more masculine men for short term relationships and more feminine men for long-term relationships. In this way, less attractive women (who may be more at risk of desertion by their partners) may be reducing the risk of desertion by long-term partners by selecting more feminine men (who may be less likely to desert their partners) Masculinity as a handicapping signal In humans, testosterone influences the growth and masculinising of facial structures in men (Thornhill & Gangestad, 1993), and male facial masculinity may be an indicator of good genes, since it is associated with increased disease resistance and 25

33 developmental stability (Thornhill & Gangestad, 2006a). Masculinity may therefore be a handicapping signal, indicating good genes (Folstad & Karter, 1992). Facial dominance, which is closely related to masculinity (Mazur & Booth, 1998) and pubertal testosterone levels (Swaddle & Reierson, 2002), may be considered an indicator of competence and social dominance (Mueller & Mazur, 1998). In a sample of American military officers, men with higher rated facial dominance were found to achieve more career success, achieving more promotions and a higher maximum rank (Mueller & Mazur, 1998). However, this was only true of men with good cognitive, social and athletic skills. For men who lacked these attributes, facial dominance was a significant hindrance to career success. Further, men with more career success had more children and grandchildren. This suggests that facial dominance is a handicapping signal. Those who have significant positive attributes reap the rewards of exhibiting facial dominance signals, whilst less endowed men pay a high cost of bearing facial dominance signals, both in terms of career advancement and in terms of reproductive success (Mueller & Mazur, 1998). 1.9 Skin Colour Though a number of different colour spaces may be used to measure and describe colour, the CIELab 1976 colour space is used throughout this thesis. This colour space is modelled upon the human visual system and is designed to be perceptually uniform (a change of one unit will appear to be of approximately the same magnitude regardless of the dimension of the change). The CIELab colour space is commonly used in human perceptual work (Martinkauppi, 2002), and is defined by L* (luminance), a* (red-green) and b* (yellow-blue) dimensions (Fig 1.1). 26

34 Fig. 1.1: A graphic representation of the CIELab colour space. From Structure and function of human skin The skin is the largest and most visible organ in the human body. It serves to protect the body from the environment, including UV radiation, and to regulate body temperature and the amount of water and other chemicals entering and leaving the body (Jablonski, 2006). The skin is made up of a number of layers, which are considered in two main components, the dermis and the epidermis (Robins, 1991). The dermis, the deeper component, is a layer of connective tissue, made up of collagen, elastin and reticular fibres. These fibres give the skin its strength and elastic properties. The dermis contains a number of structures, including hair follicles, sweat glands, sebaceous glands, arrector pili muscles (which cause the hairs to stand on end in some circumstances) and a network of lymphatic and blood vessels (Robins, 1991). The epidermis is the component of the skin closer to the surface. It is made up of four sub-layers of skin cells know as keratinocytes, which are embedded in a matrix of lipids and proteins (Elias & Feingold, 2003). Keratinocytes contain keratin filaments 27

35 embedded in a gelatinous matrix. This construction gives the epidermis its strength and durability (Jablonski, 2006). The deepest of these layers of keratinocytes is called the basal layer or stratum basale. This is a single layer of columnar keratinocytes in which mitosis (cell division) takes place, interspersed with a number of melanocytes, which produce melanin. Superficial to the basal layer, the stratum spinosum is composed of several layers of keratinocytes. Together, the stratum basale and the stratum spinosum are referred to as the Malpighian layer. Superficial to this is the stratum granulosum, which consists of several layers of keratinocytes, which gradually become flattened and die. The stratum corneum is the most superficial layer of the skin and it consists of several layers of dead, flattened keratinocytes. New keratinocytes are produced by mitosis in the stratum basale and migrate towards the surface, where they are lost from the stratum corneum by abrasion, in a process known as desquamation. The time taken for a cell to be lost from the skin surface from its production in the stratum basale is typically four to six weeks (Robins, 1991) Skin Pigmentation A number of pigments within the dermis and epidermis give the skin its colour Blood In order to keep the skin supplied with oxygen, the dermis contains a rich network of blood vessels. Arterioles bring oxygenated blood from the heart, and venules remove deoxygenated blood from the skin. Capillaries allow the diffusion of oxygen and carbon dioxide across their thin membranes, supplying cells with oxygen and removing carbon dioxide. These capillaries also join arterioles to venules. 28

36 The effect of the blood on the colour of the skin depends upon both the amount of blood present (which is dependent on the degree of vascularisation and vasodilation) and the oxygenation state of the haemoglobin in the blood (Zonios et al., 2001). Blood containing a high level of oxygenated haemoglobin has a bright red colour, impacting on the a* (red-green) dimension of the CIELab colour space (Fig 1.1), whereas blood with high levels of deoxygenated haemoglobin has a duller, bluish-red colour (Pierard, 1998), increasing a* whilst decreasing b* (yellow-blue) and L* (light-dark) and this is reflected in the skin colour (Changizi et al., 2006). Additionally, vasodilation, which occurs in order to promote heat loss (Charkoudian, 2003) and also to regulate blood pressure (Drummond & Quah, 2001), causes oxygenated arterial blood to flow into the skin, increasing the oxygenation levels of the skin blood (Liu et al., 1992) Carotenoids Carotenoids are a group of yellow-red pigments that are found in the skin (Edwards & Duntley, 1939). They are obtained from the diet, chiefly from fruit and vegetables, and cannot be synthesised de novo in the body (Alaluf et al., 2002b). Carotenoids require the presence of lipids in the diet to be absorbed (Prince & Frisoli, 1993) and are transported in the blood serum bound to lipoproteins (Krinsky et al., 1958). Carotenoid supplements have been shown to increase the levels of carotenoids in the blood serum and in the skin (Stahl et al., 1998). Skin carotenoid levels affect skin colour, contributing to the b* (yellowness) component of skin colour (Alaluf et al., 2002b) and very high carotenoid intake can lead to a harmless yellow discolouration of the skin (Monk, 1983). In Section 2 I investigate the hypothesis that natural dietary intake of carotenoids and fruit and vegetables is related to the yellowness of the skin. 29

37 Carotenoids are found in subcutaneous fat (Edwards & Duntley, 1939), and have been detected in all layers of the skin (Alaluf et al., 2002b). Carotenoid concentrations are particularly high in the stratum corneum as they are secreted through the sebaceous (oil secreting) glands and reabsorbed by the stratum corneum (Edwards & Duntley, 1939; La Placa et al., 2000). Thus the concentrations of carotenoids vary around the body, and are especially high in the palms of the hands and soles of the feet, due to their unusually thick stratum corneum. Carotenoids have a role in photoprotection in the skin. Skin samples with higher levels of carotenoids have been found to have a higher minimum erythemal dose (MED; the minimum amount of UV radiation required to cause erythema or skin reddening; Alaluf et al., 2002b), and carotenoid supplementation has been found to reduce UV radiation-induced erythema in healthy skin (Stahl et al., 2000). This mechanism is probably related to the carotenoids ability to protect against reactive oxygen species (free radicals that are produced in the body during UV irradiation or by immune defence mechanisms, and may damage molecules such as DNA or proteins; Stahl et al., 2000). Carotenoids have also been found to protect the skin from visible light in photosensitive individuals (Matthews-Roth et al., 1974) Melanin Melanin is the pigment that most people would think of when considering skin colour. Melanin is a dark brown pigment and impacts negatively on the L* dimension (darkens the skin) and positively on the b* dimension (increases skin yellowness) in the CIELab colour space (Alaluf et al., 2002c; Shriver & Parra, 2000). 30

38 Melanin is produced in melanosomes on the lamellae of skin cells called melanocytes (Billingham, 1948), before being deposited in the surrounding keratinocytes (Robins, 1991). Larger melanosomes remain as individual structures whilst smaller melanosomes are packaged together into melanosome clusters surrounded by a membrane (Toda et al., 1972). While the skin of different ethnic groups varies considerably in terms of colour, and most of this variation is attributable to melanin, the preponderance of melanocytes is similar between the ethnic groups (Alaluf et al., 2002a). The difference in colour is produced by the increased activity of melanocytes in darker skin, producing more melanin (Iozumi et al., 1993) and the relatively larger size and increased melanisation of melanosomes in darker skin (Toda et al., 1972). Additionally, melanosomes in darker skin contain more dihydroxyindole (DHI)-enriched (darker brown) eumelanin and less 5,6-dihydroxyindole-2-carboxylic acid (DHICA)-enriched (lighter brown) eumelanin and yellow-red pheomelanin than lighter skin (Alaluf et al., 2002a), though pheomelanin contributes a very small amount to skin colour, even in lightly pigmented groups (Alaluf et al., 2002a). UV light exposure causes sun tanning in most people (notably not in fair skinned, red headed individuals), which consists of both an immediate and a delayed tanning response. The immediate response involves the oxidation of existing melanin, causing it to become darker (Robins, 1991), but no increase in the number or size of melanosomes (Jimbow et al., 1974; Robins, 1991). The delayed tanning response involves the production of more melanosomes and changes in the proportions of the melanin types present. While phaeomelanin (yellow-red) increases very little, 31

39 DHICA-enriched melanin (light brown) increases rather more and DHI-enriched melanin (dark brown) increases much more markedly (Alaluf et al., 2001). Photoprotection is usually considered to be the primary role of melanin, which it achieves by scattering and absorbing UV radiation in the epidermis (Kollias et al., 1991). Melanin also acts as a stable (less reactive and therefore less damaging to cells) free radical, soaking up damaging reactive oxygen species (more reactive free radicals that cause cell damage) liberated by UV radiation (Robins, 1991). Darker skinned people suffer less DNA damage after UV irradiation than light skinned people (Tadokoro et al., 2003), and are much more resistant to UV-induced erythema and skin cancer (Robins, 1991). However, phaeomelanin releases free radicals during UV irradiation (Hill & Hill, 2000), which may partially explain why individuals, with high levels of phaeomelanin (such as red-heads) experience higher levels of skin cancer and erythema than individuals with more eumelanin, though other factors may also contribute (Hennessy et al., 2005). The melanin-producing cascade plays a significant role in the immune defence of insects (Boman & Hultmark, 1987). In humans, melanocytes have phagocytic function (engulfing invading pathogens) and melanosomes have lysosomal function (destroying invading cells, once they have been engulfed by the melanocyte; Burkhart & Burkhart, 2005). It has been suggested that the primary role of melanin is immune defence rather than photoprotection, as geographical areas with darker-skinned people have higher parasite loads as well as higher UV radiation loads (Burkhart & Burkhart, 2005). However, levels of UV radiation are much higher at equatorial latitudes than towards the pole. The skin tone of indigenous people becomes lighter away from the 32

40 equator, leading to suggestions that the reduced availability of UV light at high latitude led to the evolution of light skin (Jablonski & Chaplin, 2000) The evolution of Primate Trichromatic Colour Vision The light-sensitive cones in the retina detect different specific wavelengths of light, depending upon the specific photopigments they contain (Jacobs, 1993). In most mammals, which have dichromatic vision, the retina contains two types of cone: one that responds to short (blue) wavelengths (S cones), and one that responds to long wavelengths (L cones). This gives dichromatic, yellow-blue colour vision. The S photopigment is encoded by a gene on an autosome, whereas the L photopigment is encoded by a gene on the X chromosome (see Surridge et al., 2003). In Old World (catarrhine) primates and apes, the L gene has been duplicated on the X chromosome, and one copy has mutated to encode for a third photopigment, sensitive at intermediate wavelengths (M cones), giving trichromatic vision, in which red and green are also distinguishable (e.g. Dulai et al., 1999). In most species of New World monkeys, however, this duplication has not taken place. Instead, the L gene is polymorphic, with several different alleles encoding for L photopigments. Since the gene is X-linked, males and homozygous females are dichromatic, whereas heterozygous females are trichromatic (in heterozygous females, X inactivation causes the production of different photopigments in different cone cells; Regan et al., 2001). Different theories have been advanced to explain the selective advantage that may have caused the evolution of trichromatic colour vision in primates, and also the point in time that such an evolutionary event occurred. Explanations for the evolution of 33

41 trichromatic colour vision in primates tend to focus on increased ability to identify red and orange fruit against a background of green leaves (Allen, 1879). This advantage has been confirmed by modelling (Regan et al., 2001; Riba-Hernandez et al., 2004), though field studies have failed to find an advantage for trichromatic individuals over dichromatic individuals in finding these fruits (Dominy et al., 2003; Hiramatsu et al., 2008). Another hypothesis suggests that folivory is more important that frugivory in the evolution of trichromatic colour vision in primates. In several species of plant, leaves become lighter and greener as they get older. As these leaves age, they become tougher and lower in protein. When young, these leaves are also dappled with red. To dichromats, these leaves appear to be a similar colour to old leaves, whereas trichromats can discern the red dappling and therefore can better identify younger, more tender leaves (Lucas et al., 1998). Indeed, it has been observed that trichromatic species eat these red leaves more frequently than dichromatic species, while similar results were not found for red vs green fruits (Lucas et al., 2003). A third hypothesis that has been advanced to explain the evolution of primate trichromatic vision is the use of colour in social and sexual signalling. It has been shown that trichromatic primate species have areas of bare skin that become redder with increased dominance rank (Setchell & Dixson, 2001) and increased testosterone in male (Setchell & Dixson, 2001) and reproductive quality in female mandrills (Setchell et al., 2006). Additionally, redness is interpreted as a badge of status by other male mandrills (Setchell & Wickings, 2005) and are preferred as mates by females (Setchell, 2005). Male macaques become redder in the mating season, in 34

42 response to increased testosterone levels (Rhodes et al., 1997), and redder males are preferred by females (Waitt et al., 2003). Changes in the amount of blood in the skin and the degree of oxygenation of the blood alter the light reflectance spectrum of the skin in predictable ways. It has been shown that the wavelengths to which the long- and medium-wavelength photopigments in the cones of primates eyes respond maximally is closely matched to the parts of the spectrum that change with changing levels of blood perfusion and oxygenation. The eyes of primates with habitually trichromatic colour vision are therefore ideally suited to discerning differences in the perfusion and oxygenation of blood in the skin of conspecifics (Changizi et al., 2006). Additionally, trichromatic species are more likely to have patches of bare skin, where such changes in skin blood perfusion and oxygenation will be visible to conspecifics (Changizi et al., 2006). Humans become facially flushed with blood when angry (Drummond & Quah, 2001), when experiencing unwanted social attention (Leary et al., 1992) and in a range of social situations, including talking to someone whom the blusher finds attractive (Bogels et al., 1996), indicating a role for redness in human social and sexual signalling. It has therefore been suggested that the detection of socio-sexual signals may have provided the selective advantage for the evolution of trichromatic colour vision (Changizi et al., 2006), as being sensitive to these cues could allow an individual to avoid violent conflict with high ranking or very angry individuals, or to detect higher quality mates. Phylogenetic study has shown that primate trichromatic colour vision is likely to have evolved well before the evolution of skin and pelage redness, 35

43 suggesting that such signalling did not drive the evolution of trichromatic vision, and instead trichromatic vision enabled colour signalling to evolve (Fernandez & Morris, 2007). However, it is still possible that subtle differences in skin blood colour existed prior to the evolution of trichromatic vision. Detection of such cues could still have provided a selective advantage to those able to detect them, providing selective pressure for the evolution of trichromatic vision. The evolution of trichromatic vision, in turn, would then allow the evolution of conspicuous skin redness signals. Changizi et al. (2006) also present data showing that more routine trichromacy is associated with an increased percentage of facial bareness in the primate order, providing further support for the evolution of trichromacy to detect socio-sexual signals. Finally, while dichromatic mammals can see yellow and blue, analysis of the wavelength responses of their cones suggest that they are unable to distinguish yellow from red or green (Carroll et al., 2001). This means that signalling based on yellow pigmentation is also unlikely to be useful in such animals, as it will be easily confused with other yellow, red or green pigmentation Summary From this review of the relevant literature, it can be seen that several aspects of the human face play a significant role in attraction and mate selection, potentially signalling different aspects of mate quality, including health. The face is also the most habitually exposed skin, and contains the most social information. Skin colour varies widely both within and between populations, and is controlled by a number of pigments. These pigments have associations with health, and are used to signal condition in a number of non-human animal species. It is likely, therefore, that human 36

44 skin colour will be affected by the health of the individual, and that it will be perceived as a cue to health by observers. In the remainder of this thesis, I will examine the effect of diet on skin colour, and the effect of skin colour on the perceived health of human faces. 37

45 Section 2: Carotenoids Affect Skin Colour 38

46 2.1 Section 2 Introduction This section describes two studies examining the effect of natural daily fruit and vegetable and carotenoid intake on the colour of human skin. Study 1 shows that skin yellowness is greater in individuals that consume more fruit and vegetables and carotenoids. Study 2 uses spectral analysis to confirm that this effect is caused by increased carotenoid levels in the skin. Many species of birds and fish display brightly coloured, sexually-selected ornaments, based on carotenoids. These yellow-red ornaments are thought to signal condition and bigger, brighter carotenoid ornaments are preferred by the opposite sex. Greenfinches with brighter carotenoid-based plumage colouration have been found to have more effective immune responses, as measured by a range of metrics (Blount et al., 2003). Carotenoid levels are lower in chickens and guppies infected with parasites (Olson & Owens, 1998) and nematode infection reduces the brightness of the supra-orbital comb carotenoid ornament in red grouse (Martinez-Padilla et al., 2007), suggesting that carotenoid ornaments signal a reduced parasite load. Females of the livebearing fish Poecilia parae prefer to mate with males with carotenoid colouration (red and yellow males) than with female-coloured, dullcoloured or blue males, all of which live in the same populations (Bourne et al, 2003). Female house finches prefer to associate with males with brighter carotenoid pigmentation (Hill, 1990). Yellow-eyed penguins mate assortatively by the brightness of their yellow, carotenoid-based eye flashes, suggesting that individuals of both sexes prefer mates with brighter eye flashes (Massaro et al., 2003). 39

47 Carotenoids cannot be synthesised de novo in the body, being obtained primarily from fruits and vegetables in the diet (Alaluf et al., 2002b), and are therefore a limiting factor for organisms (Goodwin, 1984). Carotenoid ornaments may signal nutritional status and, by extension, foraging ability. Supplementing the diet of guppies with carotenoid-rich algae or with a carotenoid supplement was found to enhance the sexually-selected orange colouration of males (Karino & Haijima, 2004). In red grouse, a period of restricted feeding (thereby reducing nutritional intake and carotenoid intake) was found to reduce the redness of carotenoid ornaments (Perez- Rodrigues & Vinuela, 2008). Further, Perez-Rodriguez & Vinuela (2008) found that the carotenoid colouration of exposed skin (eye ring) was reduced more quickly than that of the beak. It is likely, therefore, that the increased yellow colouration associated with carotenoids in human skin will increase with increased dietary intake of carotenoids and will be perceived as healthier and more attractive. In this section, I investigate the relationship between natural dietary carotenoid intake and human skin colour. In later Sections, I will investigate the effects of skin colour and pigmentation on the perceived health of faces. 40

48 2.2 Study 1 Natural dietary carotenoid intake contributes to skin yellowness Introduction As discussed in Section 1, carotenoids are present in human skin, particularly in the stratum corneum. Though the levels of carotenoids in human skin are small (Robins, 1991), levels of carotenoids in the skin correlate with skin yellowness (CIELab b* values), and are therefore thought to contribute to normal human skin colour (Alaluf et al., 2002b). Additionally, supplementation of the diets of individuals with Betatene, a carotenoid-rich algal extract, over a period of 12 weeks increased levels of carotenoids in the blood serum (measured from blood samples). Measurements of skin carotenoid levels were made using reflectance spectrophotometry, and found to closely track blood serum values, also increasing with supplementation (Stahl et al., 1998). Unusually high levels of carotenoids can occur in the body, either due to excessive intake or pathological mechanisms such as a failure to convert carotenoids to vitamin A. This can lead to skin yellowing, and is known as carotenaemia (Monk, 1983). Alaluf et al. (2002) found that levels of carotenoids in the skin correlate with skin yellowness (CIELab b*), and suggested that this was attributable to normal dietary intake. However, Alaluf et al. (2002b) did not examine dietary intake of carotenoids and no direct connection from natural dietary intake of carotenoids to skin yellowness (b*) has yet been established. 41

49 : Predictions Carotenoids cannot be synthesised de novo in the body, and are obtained from the diet, particularly from fruit and vegetables (Alaluf et al., 2002b). It is hypothesised therefore that normal dietary carotenoid intake will contribute to normal human skin colour. Those individuals with higher carotenoid intake are likely to have yellower skin (as measured by the CIELab b* component). Further, this increased skin yellowing is likely to be more detectable in more photoprotected areas where melanin pigmentation (which impacts on the L* and b* components, and may therefore interfere with yellowness measurements, Stamatas et al., 2004, is less pronounced, Alaluf et al., 2002c). The palm of the hand is likely to be a region that is particularly well suited to detecting the hypothesised relationship, due to its low melanin content and thick stratum corneum (Edwards & Duntley, 1939). High intake of fruit and vegetables may be expected to be associated with other aspects of a healthy lifestyle, such as taking more exercise. Individuals who take more exercise may also be expected to be more tanned, due to more time spent out of doors, exposed to the sun. The amount of exercise taken by participants will therefore be measured controlled Methods 82 Caucasian participants (aged 18 26, 48 female, all of whom reported not having used artificial tanning products in the preceeding month) completed a diet questionnaire (Margetts et al., 1989) from which daily intake frequency for food items was calculated. The daily intake frequencies of the fruit and vegetable items from the questionnaire were summed for each participant. This gave a number of fruit and vegetable portions consumed per day for each participant. 76 (43 female) of these participants completed the Godin Leisure-Time Exercise Questionnaire. The Weekly Leisure Activity Score (WLAS) was calculated, giving a validated estimate of 42

50 exercise frequency (Godin & Shephard, 1997). 22 further participants (aged 19 28; 11 female) were shown examples of standard serving sizes of fruits and vegetables (NHS, UK) and asked the number of portions (or fraction) they ate at a sitting. From these, the mean portion size per food was calculated, and used as an estimate for all participants. Item frequency was multiplied by mean portion size consumed to obtain an estimate of daily intake per food for the 82 participants. β-carotene content for each food item was retrieved from the Canadian Nutrient File (version 2007b). These values were multiplied by the total amount of β-carotene obtained from each food item by each participant. For each participant, the β-carotene intake was summed across all food items to give a total natural daily intake of β-carotene for each participant. Fruit and vegetable, and β-carotene intake was summed across items. A Konika Minolta CM2600d reflectance spectrophotometer was used to obtain measurements of skin colour (in CIELab colour space; see section 1.9) from the left outer shoulder, outer upper arm, inner upper arm, and ventral interosseous region of the palm of the hand. The CIELab colour space was used in preference to the RGB or other alternative colour spaces because the CIELab colour space is modelled on the human visual system, and is perceptually uniform (a change of one unit in CIELab colour space is of the same magnitude perceptually regardless of the initial colour and regardless of the dimension in which the change was made; see section 1.9) : Statistical Methods Some dietary variables were not normally distributed (Kolmogorov-Smirnov tests p<0.05), therefore non-parametric statistical tests were used in the analysis. Correlations (Spearman s ρ) were used to test for a relationship between daily intake of fruit and vegetable portions and skin yellowness (b*) and also between daily β- 43

51 carotene intake and skin yellowness (b*). Non-parametric partial correlations (Spearman s ρ) were used to test for these relationships, whilst controlling for the amount of exercise engaged in by the participants (WLAS) Results The palm of the hand is a skin area that shows carotenoid colour well (Stahl et al., 1998), and is subject to minimal melanisation. I found that palm skin yellowness (b*) correlated with dietary intake of fruit and vegetables (ρ=0.451; p<0.001; n=82, Fig. 2.1A), and with estimated natural β-carotene intake (ρ=0.471; p<0.001; n=82, Fig. 2.1B). This relationship was apparent at other skin locations, but was strongest in skin regions that are not habitually exposed to sunlight (Table 2.1). Such topographic variation is expected because melanisation, which also affects yellowness, would degrade but not eliminate correlations between dietary carotenoid and skin yellowness. Though a relationship was found between exercise level (WLAS) and the natural daily intakes of fruit and vegetables (ρ=0.328; p=0.002; n=76) and β-carotene (ρ=0.286; p=0.012; n=76), controlling for exercise level of participants did not weaken the correlation between skin yellowness and fruit and vegetable (ρ=0.428; p<0.001; df=73) and β-carotene (ρ=0.451; p<0.001; df=73) intake. Hence the relationship between skin colour and diet does not appear to be a by-product of other aspects of a healthy lifestyle, as measured by the WLAS. I have shown here that natural dietary fruit and vegetable and β-carotene intake affects the colour of human skin, making it more yellow. 44

52 A Palm skin yellowness (b*) Daily fruit and vegetable portions B Palm skin yellowness (b*) Natural daily β-carotene intake (mg) Fig Diet and skin colour. (A) Daily fruit and vegetable intake correlates with palm skin yellowness (b*; ρ=0.451; p<0.001; n=82). Yellow zone indicates US recommended daily intake range of 5-13 servings (Institute of Medicine, 2000). (B) Natural daily β-carotene intake correlates with palm skin yellowness (b*; ρ=0.471; p<0.001; n=82). Yellow zone shows recommended intake to reduce the incidence of cancer (Dietary Guidelines Advisory Committee, 2005). Table 2.1. Relationships between daily fruit and vegetable and β-carotene intake and skin yellowness (b*) at various body locations. Spearman s rho (ρ) and probability (p) are given. Nonparametric partial correlations are used to control for exercise level. Shaded cells indicate statistically significant relationships. Fruit & vegetables (n=82) β-carotene (n=82) Daily intake Fruit & vegetables, (controlling exercise, df=73) β-carotene (controlling exercise, df=73) ρ p ρ p ρ p ρ p Shoulder Upper arm Under arm Palm < < < <0.001 Mean

53 2.2.4 Discussion My results demonstrate that skin yellowness is affected directly by natural fruit and vegetable and β-carotene intake. Further, I have shown that the palm of the hand, with its thick stratum corneum (Stamatas et al., 2004) and minimal melanisation exhibits the strongest correlation, and more photoexposed areas such as the upper arm show less strong correlations. Since carotenoids are especially prevalent in the stratum corneum (Edwards & Duntley, 1939), and the yellow carotenoid signal is expected to be degraded by melanisation (Alaluf et al., 2002c), this adds weight to the argument that it is the carotenoid content of the diet that is contributing to skin yellowness. Further, since controlling for exercise levels did not remove the relationship of fruit and vegetable and β-carotene intake with skin yellowness, it is unlikely that other lifestyle factors explain the relationship. Palm skin yellowness thus has potential as an accessible biomarker of diet composition similar in power to serum β-carotene measurement (skin yellowness ρ=0.45 in this study; serum β-carotene measurement r=0.48, Polsinelli et al., 1998). I have shown that skin yellowness is a reliable indicator of a healthy diet, high in fruit and vegetables. It may be hypothesised, therefore, that skin yellowness may provide a perceptible cue to health. 46

54 2.3 Study 2 Reflectance Spectra Introduction In Study 1, I showed that skin yellowness is predicted by natural dietary intake of fruit and vegetables and carotenoids. Further, since controlling for exercise did not remove the correlation, it is unlikely that the relationship is attributable to other lifestyle factors. In this section, I will use scanning reflectance spectroscopy to confirm that the relationship between skin yellowness and natural dietary intake of fruit and vegetables and carotenoids is caused by dietary carotenoids in the skin. It is possible to detect carotenoids in the skin by non-invasive reflectance spectrophotometry (Stahl et al., 1998). This method compares the reflectance of light from the skin at different wavelengths to the reflectance spectra of skin pigments to determine the pigment content of the skin. These methods have shown that carotenoid supplementation causes increased carotenoid levels in the skin (Stahl et al., 1998), and that carotenoid levels in the blood serum are correlated with levels in the skin (Stahl et al., 1998). Excessive intake of carotenoids, and certain metabolic disorders can lead to an abnormal but harmless yellowing of the skin (Monk, 1983). Further, Alaluf et al. (2002b) found that levels of carotenoids in the skin correlate with skin yellowness (CIELab b*), and suggested that this was attributable to normal dietary intake. However, Alaluf et al. (2002b) did not examine their participants diets, and the connection of natural dietary intake of carotenoids to skin colour has not yet been reliably established. 47

55 The different pigments in the skin have different absorption and reflectance spectra, absorbing light at certain wavelengths and reflecting light at others. It is these unique spectra that give the pigments their unique colours. In the visible spectrum, carotenoid absorption is best seen in the range nm (Miller, 1937; Stahl et al., 1998). It is this absorption in the short wavelength (violet, blue, green) part of the spectrum and reflectance at longer wavelengths (green, yellow, red) that gives carotenoids their yellow-red colour. β-carotene absorption peaks at 453nm (in the violet range), and 480nm (blue), α- carotene absorption peaks at 447nm (violet) and 475nm (blue) and lycopene absorption peaks at 447nm (violet), 473nm (blue) and 505nm (green) (Miller, 1937). In contrast, melanin shows a smooth increase in absorption towards the short wavelength (violet) end of the spectrum, peaking at 335nm (ultraviolet) (Kollias, 1995). Oxyhaemoglobin shows absorption peaks at 415nm (violet), 542nm (green) and 576nm (yellow) and deoxyhaemoglobin shows peaks at 439nm (violet) and 556nm (green) (Edwards & Duntley, 1939). Increases in the levels of carotenoids in the skin are expected, therefore, to increase the absorption, and to produce a corresponding decrease in reflectance, of light at the wavelengths associated with carotenoid absorption. In other words, a negative relationship between skin reflectance and dietary carotenoid intake should be observed at the wavelengths associated with carotenoid absorption. This effect should be strongest at the wavelengths with the greatest carotenoid absorption values. Therefore, the correlation coefficients should themselves be negatively related to the absorption spectra of carotenoids. 48

56 2.3.2 Methods Thirty-seven participants (29 female, 8 male) completed food frequency questionnaires (Margetts et al., 1989), and estimated daily β-carotene intake was calculated, as described in Study 1. Spectral reflectance measurements of skin colour at seven locations on the arm, hand and face (table 2.2) were made using a Konika Minolta CM-2600d spectrophotometer, sampling every 10nm from nm (a range that encompasses the entire visual spectrum and extends into ultraviolet at the short wavelength end and infrared at the long wavelength end). To control for overall lightness of the skin, reflectance values were divided by the total reflectance for each individual. Standard absorption coefficient values for three common carotenoids (α-carotene, β- carotene and lycopene) were derived from Miller (1937). Mean absorption coefficient values at each wavelength were also computed to give an estimate of a typical carotenoid absorption spectrum. Melanin absorption coefficient values were derived from (Sarna & Swartz, 1988), and haemoglobin values were obtained from Scott Prahl at omlc.ogi.edu Statistical Methods and Results Since some variables were non-parametrically distributed, Spearman s rank correlations were used for data analyses. To test for the presence of a carotenoid signal, Spearman s rank correlations were used to test for relationships between daily β-carotene intake and skin reflectance values at the wavelengths associated with carotenoid absorption ( nm). Significant relationships were found at 49

57 wavelengths corresponding to peak carotenoid absorption coefficients (table 2.2). These correlation coefficients were plotted and compared to carotenoid absorption coefficients (fig. 2.2). To test for relationships between the curves plotted, Spearman s rank correlations were used. Significant correlations were found, particularly for β-carotene and lycopene, and for the mean signals (table 2.3). Similar results for the relationship between fruit and vegetable intake and skin reflectance were also found (tables ) Examination of the plots of correlation coefficients overlaid with the absorption coefficients of common carotenoids and other common skin pigments strongly supports the connection of natural daily β-carotene and fruit and vegetable intake with skin colour (figs ). 50

58 Inner Arm 400 ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p= ρ= p=0.442 Right Cheek ρ=0.059 p=0.730 ρ=0.028 p=0.867 ρ=0.064 p=0.709 ρ=0.053 p=0.756 ρ=0.028 p=0.868 ρ= p=0.959 ρ=0.008 p=0.963 ρ= p=0.814 ρ= p=0.601 ρ= p=0.438 ρ= p=0.574 ρ=0.062 p=0.715 ρ=0.292 p=0.080 ρ=0.346 p=0.036 ρ=0.344 p=0.037 Left Cheek ρ= p=0.901 ρ= p=0.976 ρ=0.004 p=0.983 ρ=0.007 p=0.967 ρ= p=0.887 ρ= p=0.868 ρ= p=0.659 ρ= p=0.545 ρ= p=0.357 ρ= p=0.297 ρ= p=0.545 ρ= p=0.866 ρ=0.246 p=0.142 ρ=0.309 p=0.063 ρ=0.240 p=0.152 Forehead ρ= p=0.160 ρ= p=0.132 ρ= p=0.170 ρ= p=0.137 ρ= p=0.072 ρ= p=0.043 ρ= p=0.036 ρ= p=0.054 ρ= p=0.036 ρ= p=0.039 ρ= p=0.058 ρ= p=0.185 ρ= p=0.746 ρ= p=0.895 ρ=0.010 p=0.953 Upper Arm ρ= p=0.131 ρ= p=0.134 ρ= p=0.142 ρ= p=0.135 ρ= p=0.097 ρ= p=0.090 ρ= p=0.082 ρ= p=0.076 ρ= p=0.068 ρ= p=0.056 ρ= p=0.053 ρ= p=0.087 ρ= p=0.170 ρ= p=0.264 ρ= p=0.308 Shoulder ρ= p=0.044 ρ= p=0.044 ρ= p=0.047 ρ= p=0.041 ρ= p=0.037 ρ= p=0.022 ρ= p=0.016 ρ= p=0.012 ρ= p=0.010 ρ= p=0.008 ρ= p=0.009 ρ= p=0.019 ρ= p=0.023 ρ= p=0.037 ρ= p=0.073 Table 2.2: Correlations between natural daily β-carotene intake and reflectance at each wavelength, for different body locations. Significant relationships highlighted in dark grey. Marginally significant relationships highlighted in light grey. Palm ρ= p=0.797 ρ= p=0.505 ρ= p=0.299 ρ= p=0.180 ρ= p=0.111 ρ= p=0.060 ρ= p=0.053 ρ= p=0.077 ρ= p=0.050 ρ= p=0.027 ρ= p=0.036 ρ= p=0.095 ρ= p=0.248 ρ= p=0.431 ρ= p=0.560 β- carotene α- carotene Lycopene Mean Inner Arm ρ= p=0.006 ρ= p=0.004 ρ= p=0.265 ρ= p=0.013 Right Cheek ρ= p=0.003 ρ= p=0.023 ρ= p=0.001 ρ= p=0.002 Left Cheek ρ= p=0.015 ρ= p=0.117 ρ= p<0.001 ρ= p=0.010 Forehead ρ= p<0.001 ρ= p=0.002 ρ= p=0.002 ρ= p<0.001 Upper Arm ρ= p=0.024 ρ= p=0.148 ρ= p<0.001 ρ= p=0.012 Shoulder ρ= p=0.137 ρ= p=0.529 ρ= p<0.001 ρ= p=0.066 Table 2.3: Correlations between correlation coefficients given in table 2.2 and absorption coefficients of common carotenoids, for different body locations. Significant relationships highlighted in grey. Palm ρ= p=0.017 ρ= p=0.154 ρ= p<0.001 ρ= p=

59 Inner Arm 400 r= p= = p= = p= = p= = p= = p= = p= = p= = p= = p= = p= = p= = p= = p= = p=0.333 Right Cheek =0.044 p=0.794 = p=0.966 =0.007 p=0.963 =0.020 p=0.905 =0.009 p=0.955 = p=0.819 = p=0.894 = p=0.664 = p=0.533 = p=0.378 =-0.11 p=0.501 =0.016 p=0.923 =0.226 p=0.177 =0.284 p=0.088 =0.298 p=0.072 Left Cheek = p=0.756 = p=0.745 = p=0.748 = p=0.819 = p=0.618 = p=0.584 = p=0.406 = p=0.312 = p=0.193 = p=0.132 = p=0.290 = p=0.525 =0.148 p=0.381 =0.207 p=0.218 =0.162 p=0.337 Forehead = p=0.117 = p=0.059 = p=0.066 = p=0.061 = p=0.073 = p=0.083 = p=0.056 = p=0.082 = p=0.060 = p=0.092 = p=0.157 = p=0.356 =0.024 p=0.885 =0.077 p=0.642 =0.106 p=0.523 Upper Arm = p=0.086 = p=0.094 = p=0.091 = p=0.087 = p=0.060 = p=0.056 = p=0.047 = p=0.044 = p=0.041 = p=0.034 = p=0.033 = p=0.051 = p=0.101 = p=0.151 = p=0.185 Shoulder = p=0.055 = p=0.050 = p=0.061 = p=0.055 = p=0.050 = p=0.033 = p=0.025 = p=0.019 = p=0.018 = p=0.015 = p=0.015 = p=0.026 = p=0.029 = p=0.039 = p=0.083 Table 2.4: Correlations between natural daily fruit and vegatable intake and reflectance at each wavelength, for different body locations. Significant relationships highlighted in dark grey. Marginally significant relationships highlighted in light grey. Palm = p=0.704 = p=0.471 = p=0.282 = p=0.178 = p=0.149 = p=0.091 = p=0.077 = p=0.122 = p=0.117 = p=0.066 = p=0.050 = p=0.118 = p=0.178 = p=0.261 = p=0.354 β- carotene α- carotene Lycopene Mean Inner Arm r= p=0.016 r= p=0.008 r= p=0.277 r= p=0.022 Right Cheek r= p=0.005 r= p=0.041 r= p=0.001 r= p=0.003 Left Cheek r= p=0.012 r= p=0.101 r= p<0.001 r= p=0.007 Forehead r= p=0.001 r= p=0.002 r= p=0.221 r= p=0.004 Upper Arm r= p=0.020 r= p=0.136 r= p<0.001 r= p=0.009 Shoulder r= p=0.219 r= p=0.668 r= p<0.001 r= p=0.111 Table 2.5: Correlations between correlation coefficients given in table 2.4 and absorption coefficients of common carotenoids, for different body locations. Significant relationships highlighted in grey. Palm r= p=0.052 r= p=0.313 r= p<0.001 r= p=

60 Absorption coefficient [cm-1 (mg/ml)-1] Wavelength (nm) Correlation coefficient (ρ) Fig. 2.2: Correlation coefficients (solid black line) of the relationship between dietary β-carotene intake and skin reflectance values at 10nm intervals, at the palm of the hand. Similar graphs for the other skin areas can be found in Appendix A. Dashed lines show absorption spectra for common carotenoids. Red triangles=lycopene, blue rhombi=β-carotene, purple squares=α-carotene. Black dashed line shows mean absorption spectrum for the three carotenoids. 53

61 Absorption coefficient [cm-1 (mg/ml)-1] Correlation coefficients melanin Wavelength (nm) Correlation coefficient (ρ) Fig. 2.3: Correlation coefficients (solid black line) of the relationship between dietary β-carotene intake and skin reflectance values at 10nm intervals, at the palm of the hand. Brown line (circular markers) shows absorption spectrum for melanin. 54

62 Molar extinction coefficient [cm-1 (mol/ml)-1] Wavelength (nm) Correlation coefficient (ρ) Fig. 2.4: Correlation coefficients (solid black line) of the relationship between dietary β-carotene intake and skin reflectance values at 10nm intervals, at the palm of the hand. Dashed lines show absorption spectra for oxygenated (red rhombi) and deoxygenated (blue squares) haemoglobin (from Scott Prahl, omlc.ogi.edu). Green dashed line (triangular markers) shows absorption difference between oxygenated and deoxygenated blood (deoxygenated minus oxygenated blood) 55

63 Absorption coefficient [cm-1 (mg/ml)-1] Wavelength (nm) Correlation coefficient (ρ) Fig. 2.5: Correlation coefficients (solid black line) of the relationship between dietary fruit and vegetable intake and skin reflectance values at 10nm intervals, at the palm of the hand. Similar graphs for the other skin areas can be found in Appendix A. Dashed lines show absorption spectra for common carotenoids. Red triangles=lycopene, blue rhombi=β-carotene, purple squares=α-carotene. Black dashed line shows mean absorption spectrum for the three carotenoids Discussion The relationship between the natural daily β-carotene intake and the reflectance at each wavelength in the relevant range is highly similar to the inverse of the absorption spectra of common carotenoids. This contrasts with the lack of similarity between the correlation curve and the absorption spectra of the other major yellow pigment, melanin and also from the curves of oxygenated and deoxygenated haemoglobin, and the absorption spectral change associated with a change in the ratio of oxygenated to 56

64 deoxygenated haemoglobin. This evidence provides strong support for the hypothesis that increased natural dietary intake of fruit and vegetables and carotenoid pigments affects the natural colour of people s skin. The peak absorption of light by carotenoids is in the short wavelength (violet, blue and part of green) area of the visible spectrum (Miller, 1937), giving the skin a yellowish colour. There is little similarity between the absorption spectra of other common skin pigments and the plotted correlation coefficients. This suggests that carotenoid intake, and not some other lifestyle correlate (such as exercise or time spent out of doors) is responsible for the increased yellowness of the skin of individuals with high natural daily fruit and vegetable and carotenoid intakes (see Study 1). 2.4 Section 2 Discussion Studies 1 and 2 demonstrate that individuals with higher natural dietary intake of fruit and vegetables and carotenoids have increased levels of carotenoids in the skin. These carotenoids are detectable by reflectance spectrophotometry. Further, individuals with higher natural dietary intake of fruit and vegetables and carotenoids have increased skin yellowness. Species of birds and fish display colourful, sexually selected, carotenoid based ornaments. These ornaments become larger and brighter in response to increased dietary intake of carotenoids (Karino & Haijima, 2004). This response is particularly rapid and pronounced in fleshy ornaments, such as the eye-ring skin of red grouse (as opposed to keratinised ornaments such as the beak; (Perez-Rodrigues & Vinuela, 2008). Human skin therefore has a similar relationship with nutritional factors to 57

65 those seen in species for whom carotenoid colouration is a sexually selected, condition dependant, condition signalling ornament. It is known that the carotenoid content and colour of human skin is dependant on the nutritional state of the individual. CIELab b* values are associated with carotenoid levels in the skin and serum of individuals (Alaluf et al., 2002c). I have shown that the carotenoid colour of human skin is directly dependant upon the nutritional composition of the diet. Individuals with a higher natural dietary intake of carotenoids show increased skin carotenoid levels and increased skin yellowness (CIELab b*). Given the condition-dependent nature of the carotenoid colouration of human skin, and the direct relationship between carotenoid ornament brightness and mate choice in many species of birds and fish (Blount et al., 2003; Bourne et al., 2003; Saks et al., 2003), the relationship of skin carotenoid colouration (and indeed colouration in general) to perceptions of health (and therefore attractiveness) in humans is worthy of study. The remainder of this thesis will examine the role of skin colouration in health perception in humans. 58

66 Section 3: Pure Colour Transforms 59

67 3.1 Section 3 Introduction This Section describes a suite of three experiments, each examining the effect of skin colour in one of the CIELab colour dimensions (lightness, redness and yellowness) on the apparent health of human faces, and a fourth examining how the colour contrast between the lips and the facial skin affects the apparent health of human faces. These experiments are described together in this section to allow a more cohesive treatment of the results. The colour and texture of facial skin are associated with apparent health in humans (Fink et al., 2006; Jones et al., 2004). Fink et al. (2006) measured the distribution of colour in the faces of women of different ages and applied these distributions to standard-shaped three-dimensional faces. These stimuli therefore differed only in colour distribution. When presented to participants for rating, the older faces were perceived as older, less attractive and less healthy than the younger faces, showing that colour distribution alone contains enough information to make these judgements. Jones et al. (2004) found that ratings of health of skin patches taken from face photographs correlated positively with ratings of attractiveness of the whole faces. Jones et al. (2004) also found that manipulating the colour and texture of faces to look more or less healthy while shape information remained constant changed the attractiveness of the faces. The effect that overall skin colour has on the apparent health of faces has not been previously reported. This section investigates the relationship between perceived health and skin colour by allowing participants to manipulate colour calibrated facial photographs along CIELab L* (lightness), a* (redness) and b* (yellowness) axes to optimise healthy appearance. 60

68 3.1.1 Lightness The darkness of skin is primarily dependent on the level of melanin present in the epidermis. Melanin is a dark brown pigment that impacts positively on the b* (yellow) component and negatively on the L* (lightness) component of skin colour. Darker skin contains larger, more mature (stage 3 and 4) and more individually dispersed melanocytes than lighter skin (Toda et al., 1972). Many primate groups, including chimpanzees, have light skin under dark hair. Exposed skin in these species is lightly pigmented and becomes darker with photoexposure and with increasing age (Post et al., 1975). It is often assumed, however, that the original colour of modern human skin was black (Hamilton, 1973). As migration to the savannah took place, hominins are thought to have lost their hair as part of a suite of thermoregulatory adaptations. It is suggested that dark skins evolved to protect against the high levels of UV radiation in the African savannah, in the absence of large amounts of body hair (Jablonski & Chaplin, 2000). This hypothesis is supported by the fact that dark skin is less susceptible to sunburn (symptoms of which include erythema skin reddening - and oedema - inflammation) and skin cancer than light skin, and has higher minimum erythemal doses (MED) of UV radiation (the amount of UV radiation required to cause erythema this is the standard way of expressing tendency to burn; Robins, 1991). Therefore, dark skin is important where levels of UV radiation are high, such as near the equator (Jablonski & Chaplin, 2000). The selective pressures influencing the evolution of light skin are less clear, however. As levels of UV radiation are reduced further away from the equator, the selective pressure to maintain high levels of the dark pigment melanin in the skin are reduced 61

69 (Jablonski & Chaplin, 2000). However, population genetic studies suggest that there has not been enough time since the divergence of European and Asian populations from African populations for the high frequency of certain variants of the MC1R gene (partly responsible for skin colour) present in light skinned populations to be explained by genetic drift alone. Selective pressures must have been acting to select for light skin at high latitudes (Aoki, 2002). The vitamin D hypothesis aims to provide a selective mechanism for the evolution of light skin at high latitude (Murray, 1934). Vitamin D is synthesised in the dermis (lower level) of the skin. This synthesis requires UV light. Vitamin D is required for the normal absorption of calcium from the gut and its deposition in bones (Loomis, 1967; Neer, 1975), and vitamin D deficiency leads to rickets or osteomalacia (Schultz, 1932). Individuals with rickets show symptoms including bowed legs, which impair locomotion, and a deformed pelvis (Loomis, 1967; Neer, 1975). Pelvic deformity may also impair locomotion and, in women, can prevent natural childbirth, leading to death in the absence of Caesarian section (Jablonski & Chaplin, 2000). Individuals with rickets also have weak bones which are liable to breaking, and weaker muscles (Neer, 1975). Near to the equator, levels of UV radiation are high and vitamin D can be synthesised in sufficient amounts even by individuals with very dark skin. Away from the equator, however, the increased filtering effects of the atmosphere (Park, 1932; Neer, 1975) mean that UV radiation is reduced by up to 5-fold. Individuals with dark skin can therefore be unable to synthesise sufficient vitamin D (Jablonski & Chaplin, 2000). This lack of UV radiation is compounded by the increased seasonality that is 62

70 introduced at high latitudes. At 41ºN (the Mediterranean Sea), due to changes in the angle of the sun and day length, there is a 16-fold decrease in available UV light from summer to winter (Neer, 1975). Melanin filters out large amounts of UV radiation, limiting the amount that reaches the dermis and is therefore available for the synthesis of vitamin D (Loomis, 1967). Levels of UV radiation reduce dramatically as latitude increases. It is therefore expected that selective pressure would favour individuals with lighter skin away from the equator. Combined with the selective pressure for darker skin near the equator to protect against sunburn and skin cancer, a selection gradient is expected, from very dark skin at the equator to very light skin far from the equator. This is indeed the general pattern that is observed in indigenous people around the world (Jablonski & Chaplin, 2000). Objections to the vitamin D hypothesis have been voiced, however. Robins (1991) points out that rickets are associated less with latitude and more with behavioural patterns, such as urbanisation, spending large amounts of time indoors and wearing clothes that expose very little of the skin to the sun. These behaviours prevent UV light from reaching the skin. In the Asian community in the UK, the phytate content of chapatti flour has been suggested to inhibit the absorption of calcium in the intestine, leading to the relatively higher levels of rickets that are observed in this population (Goswami et al., 2000; Shaw, 2003). Further, though dark skinned individuals cannot synthesise vitamin D throughout the year at high latitude (Jablonski & Chaplin, 2000), vitamin D can be stored in the fat and muscle tissue 63

71 (Rosentreich et al., 1971) and over the whole year, UV levels are sufficient to produce enough vitamin D even at high latitudes (Beadle, 1977). Another hypothesis for the evolution of light skin at high latitude is sexual selection. This hypothesis suggests that sexual selection for light skin provides the selection pressure for light skin to evolve away from the equator (Aoki, 2002; Darwin, 1871). Van den Berghe & Frost (1986) reviewed evidence from the Human Relations Area File and found that, out of 51 societies that cited skin colour as a component of attractiveness, 47 expressed a preference for lighter skin. Of these, 30 mentioned light skin in relation to women only and 14 in relation to both sexes. American college students have also been found to prefer the lighter categories when asked to express a preference for descriptions of skin colour (Feinmann & Gill, 1978), though using descriptions is a questionable methodology. A number of hypotheses have been advanced to explain the preference for light skinned females. In most populations, women have lighter skin than men (Hulse, 1967; van den Berghe & Frost, 1986), even in cases where women spend more time in the sun than men (Byard, 1981). Since femininity is strongly preferred in female faces (Perrett et al., 1998), an exaggeration of this aspect of femininity may also be preferred as a feminine trait. Some evidence suggests that women prefer femininity in male faces (Perrett et al., 1998), particularly when the observer is in the luteal (infertile) phase of the menstrual cycle (Penton-Voak & Perrett, 2000; Penton-Voak et al., 1999) and for long-term relationships (Little et al., 2002), so lightness could also be preferred in male faces. 64

72 Lightness of skin may be associated with fecundity. The sexual dimorphism present in skin lightness appears at puberty, when other aspects of sexual dimorphism become more pronounced (van den Berghe & Frost, 1986). Skin darkens during pregnancy and during the infertile phase of the menstrual cycle (van den Berghe & Frost, 1986). Areolar skin also darkens during pregnancy and, after approximately three births, fails to return to its previous colour (Pawson & Petrakis, 1975). Men who are sensitive to these shifts in colour, and prefer lighter skinned women may have a reproductive advantage. However, since women become lighter with senescence, and are darkest during the reproductive years (Jablonski, 2006), these cues may need to be interpreted in conjunction with cues to youth, such as smooth, unwrinkled skin in order to provide useful cues to fecundity. Skin lightness is also associated with higher socioeconomic status in many societies (Hulse, 1967; Jones, 2000; van den Berghe & Frost, 1986) and is considered more attractive in racially stratified societies societies where wealth, health and power are associated with white skin (Jones, 2000). People may also take skin lightness or darkness as a sign of health, lightness indicating an ability to produce sufficient vitamin D in high latitude areas, and darker skin indicating increased resistance to the negative effects of UV radiation. In this case, it would be expected that lighter skin would be preferred at high latitude areas with low levels of UV radiation, and darker skin would be preferred near the equator where UV radiation levels are high. However, there is evidence that light skin is preferred even close to the equator where dark skin would be advantageous (van den 65

73 Berghe & Frost, 1986). Section 6 will investigate the affect of ethnicity on colour preference. Following previous studies, I hypothesise that light skin will be preferred by participants, particularly in female faces. A preference for averageness in faces (Langlois & Roggman, 1990) may limit the preference for extremely light faces Redness Skin redness is affected by the amount of blood in the skin, and also by the oxygenation state of the blood (Zonios et al., 2001), with oxygenated blood having a brighter red colour and deoxygenated blood having a darker, bluer red colour (Pierard, 1998). Increased facial redness is associated with higher dominance rank and increased testosterone levels in male non-human old world primates (that have trichromatic colour vision; Rhodes et al., 1997; Setchell & Dixson, 2001), and is preferred by females (Waitt et al., 2003). In female mandrills, facial redness varies across the menstrual cycle, brightening during the follicular (fertile) phase of the menstrual cycle and also with maturity (Setchell et al., 2006). Sexual skins such as the perineum, genitalia and surrounding areas also signal oestrus in several species of primate (Dixson, 1983). Indeed, it has been suggested that primate colour vision evolved to allow the detection of socio-sexual signals based on skin blood perfusion and oxygenation (Changizi et al., 2006). Increased skin blood flow is also associated with physical fitness (see Section 4 for a more detailed discussion of blood colour). It is therefore hypothesised that increased levels of redness at rest will be considered healthier in human faces. 66

74 Sexual dimorphism in skin redness has been reported, with men being redder than women (Edwards & Duntley, 1939; Tarr et al., 2001). It may be hypothesised, therefore, that participants will enhance an existing dimorphism by reddening male faces more than female faces. Extremes of facial redness may not be preferred because of the preference for averageness (Langlois & Roggman, 1990), so it may be expected that very red faces will be reduced in redness to optimise healthy appearance Yellowness Increased skin yellowness is associated with increased levels of carotenoids (Alaluf et al., 2002b; Section 1) and melanin (Stamatas et al., 2004) in the skin. Increased carotenoid levels are associated with increased reproductive health and immunocompetence in many species, including humans. Preference for carotenoids may be expressed as a preference for yellowness in the skin. Melanin may have health benefits or costs. Preference for increased melanin will result in a preference for yellower (and darker) faces, if melanin is driving the relationship between yellowness and apparent health (see Section 5 for a more detailed discussion of carotenoids; melanin is discussed above). It is hypothesised that increased skin yellowness will appear healthy. To test the hypotheses discussed in this section, a computer programme will be used to allow participants to manipulate the colour of the skin portions of colour calibrated photographs along CIELab L* (lightness), a* (redness) and b* (yellowness) axes to optimise the healthy appearance of the faces. 67

75 3.1.4 Contrast between skin and lips In light-skinned faces, the features, especially the eyes and lips, are darker in colour than the surrounding areas. Increasing this contrast increases the attractiveness of female faces, but decreases the attractiveness of male faces (Russell, 2003). The contrast between the features and the surrounding areas is naturally greater in women than in men, and images with increased facial contrast are judged to be more feminine (female faces) or less masculine (male faces). Women s use of makeup (presumably to enhance the healthy, attractive appearance of the face) has also been found to increase this contrast (Russell, pers comm..). These studies used greyscale images in much of the work, and only considered luminance, not colour differences. Additional conditions in the current study will involve the manipulation of the facial skin both while the lips remain unchanged and while the lips are colour manipulated along with the rest of the facial skin. This will allow the investigation of the effects of the contrast between the lips and the facial skin on all three CIELab colour axes. Increasing the luminance of the facial skin while maintaining the original luminance of the lips will increase the contrast between the facial skin and the lips. If the luminance of the lips is manipulated along with the facial skin, the contrast of the skin with the lips will remain unchanged. Following Russell s hypothesis, it may be predicted that the lips constant condition will increase apparent health more than the lips manipulated condition in female faces. Male faces are expected to show the opposite pattern. Increasing the redness of the facial skin while the lips remain constant will reduce the relative redness of the lips, or even make them appear green in comparison. Further, many lipsticks used by women exaggerate the redness of the lips (perhaps as a sexual 68

76 signal that imitates the labia of the female genitalia; Low, 1979). This may restrict the amount of redness added, particularly to female faces, to optimise healthy appearance. In this case, participants are expected to increase facial redness less in the lips constant condition than the lips changing condition, particularly for female faces. Increasing yellowness of the facial skin when the lip colour remains constant may give the lips a blue appearance. In this case, more participants are expected to increase the yellowness of the facial skin less to optimise healthy appearance when the lips remain constant as the rest of the skin changes around it than when the lips change along with the rest of the facial skin. 3.2 Methods Photography 51 Caucasian participants (30 female, 21 male, aged 18 22) from the University of St Andrews were photographed with neutral expressions and were not wearing makeup. The camera was a Fujifilm FinePix S2Pro digital SLR, fitted with a Nikon 60mm fixed length lens. Photography took place within a booth, painted with a standard Munsell N5 grey paint (Verivide, UK). Illumination was provided by three d65 daylight simulation tubes fitted into high frequency fixtures (Verivide, UK) to reduce the effects of flicker on photography. A GretagMacbeth Mini ColorChecker colour chart was included in the frame. Matlab was used to read RGB values for each of the 24 colour squares on the photograph of the colour chart. A least-squares transform from an 11-expression polynomial expansion of these RGB values to manufacturer stated CIELab values of the colour patches of the colour chart was used to colour correct the photographs (Hong et al., 2001). A mean E value of 2.44 was achieved 69

77 ( E is the Euclidean distance between two points in CIELab colour space). This value is within the acceptable colour calibration limits for accurately displaying complex colour images on a cathode ray tube (CRT) monitor (Stokes et al., 1992). Additionally, displayed faces were within the colour range of the population, meaning that participants were presented with accurate and realistic images to manipulate (Appendix B). The images were delineated in Psychomorph (after Burt & Perrett, 1995). Matlab was used to read the mean CIELab L*, a* and b* values of the skin portions of each colour calibrated face to give initial CIELab values Colour Transformations Face-shaped masks were created representing a neutral skin tone, one with increased a* and one with decreased a*. The masks were delineated in Psychomorph, marking the same points as on each of the images. The masks were then warped to the shape of each facial image, and the images manipulated by the colour difference between the two masks (fig. 3.1; Burt & Perrett, 1995). This gave a series of 13 images from +16 to -16 units of a*. This process was repeated for b* and L*. Only the facial skin was manipulated. Eyes, hair, clothing and background remained constant. Because the colour masks used for the transform were even in colour distribution, no change in colour distribution occurred on the images. Two sets were produced one with the lips remaining constant and one where they changed along with the facial skin (Fig ). 70

78 a b c d e Fig. 3.1: delineated images and masks. The original image (c) is transformed by the difference between the two colour masks (b) and (d), giving facial photographs with redder (e) or greener (a) skin portions. b c a d e Fig. 3.2: a* (red-green) colour transforms. Original image (a) and images with skin portions transformed by -16 (b) and +16 (c) units of a* with lips remaining constant and with lips manipulated along with the rest of the face (d, e). 71

79 b c a d e Fig. 3.3: b* (yellow-blue) colour transforms. Original image (a) and images with skin portions transformed by -16 (b) and +16 (c) units of b* with lips remaining constant and with lips manipulated along with the rest of the face (d, e). 72

80 b c a d e Fig. 3.4: L* (light-dark) colour transforms. Original image (a) and images with skin portions transformed by -16 (b) and +16 (c) units of L* with lips remaining constant and with lips manipulated along with the rest of the face (d, e) Experimentation 32 Caucasian participants (12 male, 20 female, aged 19 26) from the University of St Andrews were presented with the stimuli on a CRT, colour calibrated using a ColorVision Spyder 2Pro to mean E of for skin tones. A computer program allowed them to manipulate the colour of the facial skin to achieve optimum healthy appearance. Moving the mouse horizontally across the screen changed the colour of the face. Participants were presented with the 51 faces, one at a time and asked to make the face as HEALTHY as possible. Participations manipulated the colour of 73

81 the face until healthy appearance was optimised and then clicked the mouse to move on to the next face. Male and female faces were presented in separate sets of trials (blocks). Order of presentation of faces and manipulations (L*, a*or b* dimension; with or without lips) were randomised within these blocks. Each participant was randomly assigned the lips manipulated or lips constant version of each face (no participant saw both versions of a face). Each participant was presented with 306 trials Statistical Methods Mean colour changes applied to each face were calculated (by face dataset). Mean colour changes applied by each participant were also calculated (by participant dataset). In the by face dataset, one-sample t-tests (H 0 : no colour change) were used to test whether the faces were changed in colour to optimise healthy appearance. Univariate ANOVAs were used to test for the effect of face sex on colour change, as this test allows initial face colour to be controlled as a covariate (by face dataset; dependent variable=colour change applied; fixed factor=face sex; covariate=initial face colour: L*, a* or b*). Pearson s r was used to test for correlation between initial face colour and colour change applied. Independent samples t-tests were used in the by participant dataset to test for differences in amount of colour change applied by male and female participants. To test for an interaction between sex of face and sex of participant, repeated-measures ANOVAs were used (dependent variable=colour change applied; within-subjects factor=participant sex; between-subjects factor=face sex). When one or more variable was not normally distributed, Mann-Whitney U tests were used in place of t-tests. Paired-samples t-tests were used to test for differences between lips changing and lips constant conditions. 74

82 3.3 Results Relative Lip Colour No difference in amount of colour transform applied to optimise healthy appearance was found between the two lip conditions (lips changing with the rest of the facial skin and lips remaining constant as the rest of the skin changes) for the luminance (L*; paired samples t test t 50 =0.187, p=0.852) or redness (a*; t 50 =0.855, p=0.397) transforms. However, participants increased the yellowness (b*) of faces more when the lips changed with the rest of the facial skin than when the lips remained constant (t 50 =2.793; p=0.007). Repeated measures ANOVAs showed no interaction between sex of face and lip condition affecting the amount of L* (F 1,49 =0.025, p=0.876), a* (F 1,49 =0.438, p=0.511) or b* (F 1,49 =0.525, 0.472) change applied to optimise healthy appearance, suggesting that the effect of lip condition on amount of colour change applied was not affected by the sex of the faces. The results of the analyses described above were not qualitatively different for the lips constant and lips changing conditions (i.e. direction and significance of analyses were not qualitatively different when repeated for each of the two conditions). Results were therefore pooled for the two conditions (lips manipulated or not manipulated) in the analyses below Lightness To optimize healthy appearance, faces were lightened in the L* transform ( L*= , t 50 =5.200, p<0.001), and the amount of lightening applied was negatively related to the initial lightness (L*) of the face (Fig. 3.5A, R 2 =0.50; Table 3.1). 75

83 Colour axis L* a* b* Correlation with initial facial colour (n=51) L* a* b* r=-0.708; p<0.001 r=0.242; p=0.087 r=-0.116; p=0.419 r=0.231; r=-0.906; p<0.001 r=0.010; p=0.945 r=-0.509; p<0.001 r=0.125; p=0.381 r=-0.772; p<0.001 Table 3.1. Correlations (Pearson s r) of colour transform with initial facial colour. Female faces were initially lighter than male faces (van den Berghe & Frost, 1986; female L*= ; male L*= ; t 49 =2.342; p=0.023). Controlling for initial facial lightness, female faces were lightened more than male faces (F 1,48 =27.398; p<0.001; Fig. 3.5B). Thus participants lightened the faces of both sexes, but increased an existing sexual dimorphism (Perrett et al., 1998) to optimise healthy appearance by lightening female faces more than male faces. A Lightness change (ΔL*) Initial lightness of face (L*) B Lightness change (ΔL*) 7 0 male faces female faces Initial facial lightness (L*) Fig. 3.5: (A) Amount of lightness change applied to optimise healthy appearance. (B) Female faces (solid symbols) are lightened more than male faces (hollow symbols) for a given value of initial face lightness (L*). Each point represents one face. Error bars show standard error of the mean across all participants. 76

84 A non-significant trend suggested that male participants may prefer lighter skin than female participants (independent samples t test; t 30 =2.032; p=0.051), suggesting a greater preference for feminine light skin by men than by women. However, this preference was not limited to female faces, as no interaction between the sex of the face and the sex of the participants was found to affect the amount of lightness change applied to optimise healthy appearance (F 1,49 =0.005; p=0.941) Redness Participants increased the redness of faces to optimise healthy appearance (Fig. 3.6A; a*= ; t 50 =8.714, p<0.001). The initial redness of the face was negatively related to the amount of redness applied to optimise healthy appearance, with initially less red faces increased in redness more than initially redder faces (r=-0.906; p<0.001; Fig 3.6A; Table 3.1). A Redness change (Δa*) Initial facial redness (a*) B Redness change (Δa*) 5 0 male faces female faces Initial facial redness (a*) Fig. 3.6: (A) Amount of redness change applied to optimise healthy appearance. (B) Male faces (hollow symbols) are increased in redness more than female faces (solid symbols) for given values of initial face redness (a*). 77

85 Male faces showed a trend to be initially redder than female faces (male a*= ; female: a*= ; t 49 =1.765, p=0.084) and were significantly redder than female faces after participants adjusted a* (male: a*= ; female: a*= ; t 49 =4.422; p<0.001). Controlling for initial colour of the face, male faces were increased in redness more than female faces (Fig. 3.6B; F 1,48 =15.708; p<0.001). Thus redness appears healthy for both male and female faces but participants enhance the natural sexual dimorphism in face colour when judging optimal appearance; participants associate a higher level of redness with optimal health in men than in women. The sex of the participants was not found to affect the amount of redness change applied to optimise healthy appearance (Mann-Whitney U test; U=93; p=0.307; n=32). No interaction between the sex of the face and the sex of the participants was found to affect the amount of redness change applied to optimise healthy appearance (F 1,49 =0.067; p=0.798) Yellowness Participants increased the yellowness of faces to optimise healthy appearance (Fig. 3.7 b*= ; t 50 =29.639, p<0.001). The increase in yellowness was negatively related to the starting yellowness (b*) of the face (Fig. 3.7, R 2 =0.60; Table 3.1). Faces initially low in yellow were altered more than faces initially high in yellow. 78

86 Yellowness change (Δb*) Initial facial yellowness (b*) Fig. 3.7: Amount of yellowness change applied to optimise healthy appearance. Yellowness change applied correlates negatively with initial face yellowness (b*). Male and female faces did not differ in initial yellowness (t 49 =1.079; p=0.286) and sex of face did not affect yellowness change applied to optimise healthy appearance (F 1,48 =0.382; p=0.540). The sex of the participants was not found to affect the amount of yellowness change applied to optimise healthy appearance (t 30 =0.690; p=0.495). No interaction between the sex of the face and the sex of the participants was found to affect the amount of yellowness change applied to optimise healthy appearance (F 1,49 =0.676; p=0.415) dimensions. 3.4 Section 3 Discussion Participants increased red (a*), yellow (b*) and lightness (L*) to optimise the healthy appearance of faces. 79

87 3.4.1 Contrast between Skin and Lips Participants added more yellow to the face when the lips changed than when they remained unchanged. This may be because increasing the b* component of surrounding areas will make the lips look bluer. Blue lips are a sign of cyanosis (a lack of oxygenated blood), which is associated with respiratory and cardiac illnesses (Ponsonby et al., 1997). There was no interaction of sex of face and lip condition affecting the amount of colour change applied to optimise healthy appearance. Since the contrast between the lips and the skin is affected differently depending on whether the lips also change in colour, no support was found for Russell s (2003) hypothesis that the relative luminance or relative colour of facial skin and facial features (i.e. the contrast between facial skin and the lips and eyes) affects the healthy appearance or attractiveness of a face. Russell s (2003) results may differ from mine because of his use of black and white photographs, and only three levels of contrast Lightness Skin lightness is associated with high social status in some populations (Hulse, 1967; Jones, 2000), and is useful in preventing vitamin D deficiency in populations far from the equator (Jablonski & Chaplin, 2000). It is also associated with femininity and fecundity in women (van den Berghe & Frost, 1986). In agreement with other studies (Hulse, 1967; also van den Berghe & Frost, 1986 list 32 such studies), there was also sexual dimorphism in initial and final skin lightness found in the current study, with female faces lighter than male faces. Controlling for initial lightness of face, more lightness was added to female faces. The existing sexual dimorphism in the lightness of faces was enhanced to optimise healthy appearance. The preference for lighter skin in female than male faces (the optimal lightness of female faces is higher than the 80

88 optimal lightness of male faces) may indicate an association between lightness, femininity and sexual dimorphism Redness The a* dimension is associated with haemoglobin levels, and the present results are consistent with the hypothesis that oxygenated haemoglobin and increased skin blood flow look healthy because they are associated with athletic health (Armstrong & Welsman, 2001) and physical training (Johnson, 1998), and with increased levels of reproductive hormones (Rhodes et al., 1997). The increase in redness to optimise healthy appearance is also consistent with the observation that reduction in skin blood flow is caused by illness and health factors such as diabetes (Charkoudian, 2003), smoking (Richardson, 1987) and hypertension (Panza et al., 1990). A more detailed examination of the role of blood colour in the perception of health in human faces is warranted and can be found in Section 4. A trend was found in the current study to suggest that male faces are initially redder than female faces. Controlling for initial redness of face, participants reddened male faces more than female faces to optimise healthy appearance. The optimal redness of male faces was higher than for female faces. Participants therefore enhanced the existing sexual dimorphism of redness in faces to optimise healthy appearance. These enhancements of sexual dimorphism indicate a role of skin colour in sexual selection (Andersson, 1994). 81

89 3.4.4 Yellowness Increased skin yellowness (b*) increases apparent health, but decreased skin lightness (low L*) does not. It is likely that increased yellowness appears healthy because of its association with increased carotenoid (yellow; increased b*) levels rather than with increased melanin levels (dark and yellow; increased b*, decreased L*). This conclusion is consistent with the evidence that carotenoids are associated with improved reproductive health (Brief & Chew, 1985) and immunocompetence, including increased number and activity of T lymphocytes (Alexander et al., 1985; Chew, 1993). It is also consistent with the fact that illness, including malaria and HIV infection is associated with low carotenoid levels (Friis et al., 2001). Further consideration of the role of carotenoid and melanin in the perception if health in human faces is warranted and can be found in Section Gradients The negative correlation between starting colour (starting a* or b*) with colour added ( a* or b*) supports a hypothesis that low colour looks unhealthy. Participants add colour to remedy the situation. The regression lines (that relate colour change applied to faces in order to optimise healthy appearance to initial face colour) cross the x-axis (a*), or would do so if extrapolated (b*), indicating that individual faces that are very red or very yellow will have colour removed to optimise healthy appearance. Similarly, faces that are very light are darkened to optimise healthy appearance, suggesting that there is an optimum colour range for healthy facial skin appearance. 82

90 Section 4: Blood Colour Transforms 83

91 Blood colouration provides perceptible cues to health in human faces A version of this section has been published as: Stephen, I.D., Coetzee, V., Law Smith, M.J., Perrett, D.I. (2009) Skin blood perfusion and oxygenation colour affect perceived human health. PLoS ONE, 4, e Introduction This Section describes a suite of three experiments examining the effect of blood colouration on the apparent health of human faces. The first experiment examines the effect of oxygenated blood colour, the second deoxygenated blood colour. The third experiment uses two-dimensional colour transforms to further examine the relationship of these pigment colours with apparent health. These experiments are described together to allow a more thorough and cohesive treatment and discussion of the results. There have been suggestions that reddening is considered attractive, both across the face (Fink et al. 2001), and specifically in the cheeks (Zahavi & Zahavi, 1997). In Section 3, I showed that increased facial skin redness increased the apparent health of faces. The red pigment, haemoglobin in blood is one of the main contributors to human skin colour (Zonios et al., 2001). Skin redness is caused by skin blood perfusion, controlled by vascularisation and vasodilation. In addition, blood oxygenation state 84

92 affects the colour of the skin, oxygenated blood being a bright red colour and deoxygenated blood being a darker, bluer shade of red (Pierard, 1998). This experiment investigates the pigment basis of the preference for redder facial skin by manipulating the oxygenated and deoxygenated blood colour of the facial skin. To allow the further investigation of the relationship between oxygenated and deoxygenated blood colour, additional experiments were performed to allow the manipulation of both oxygenated and deoxygenated blood colour simultaneously Blood Perfusion In women, high oestrogen levels are associated with increased vascularisation of the skin (Thornton, 2002) and increased oestrogen and progesterone are associated with an increased vasodilatory response to heating (Charkoudian et al., 1999), which causes arterialisation of the blood in the skin (Liu et al. 1992). Increased progesterone and oestrogen in the luteal phase of the menstrual cycle (Baird et al., 1997) and increased oestrogen in the follicular phase (Lipson & Ellison, 1996; Stewart et al., 1993) are associated with increased chance of embryo implantation. Increased oxygenated blood colouration may therefore be a good indicator of reproductive health in women. Physical training increases the responsiveness of the vasodilator mechanism as well as reducing the activity of the vasoconstrictor mechanism, providing improved heat loss during exercise (Johnson, 1998). Skin vasodilation is impaired in patients with type 2 diabetes mellitus (Charkoudian 2003), with senescence (Tankersley et al., 1991) and in smokers (Richardson, 1987). Hypertension (high blood pressure) is associated with decreased reactivity of the vasodilatory system (Panza et al., 1990) and causes 85

93 capillary rarefaction (the loss of capillaries in the skin; (Serne et al., 2001). A slightly reddish skin colour is expected to increase healthy appearance, as a cue to normal blood circulation (Fink et al., 2001) Blood Oxygenation Increased blood oxygenation is associated with aerobic fitness in year old boys and peak oxygen uptake (VO 2 max), positively associated with haemoglobin levels, is the most common measure of athletic fitness (Armstrong & Welsman 2001). Increased oxygenation of the blood in the skin causes reddening (Changizi et al., 2006). Increased levels of deoxygenated blood are associated with hypoxia and can lead to cyanosis (a blue tint to the skin), which is associated with several respiratory and coronary illnesses (Pierard, 1998) Primate Colour Signals Red-green colour vision in primates may have evolved to allow the interpretation of socio-sexual colour signalling, since the spectral sensitivities of components of the primate retina are matched to the spectra needed to identify skin blood concentration and oxygenation (Changizi et al., 2006). Reddening of the facial skin occurs with increased dominance rank in male mandrills (Setchell & Dixson 2001). In the mating season, male rhesus macaques experience facial reddening associated with increased testosterone (Rhodes et al 1997). This colour is preferred by female macaques (Waitt et al., 2003). Male macaques have also been found to prefer red colouration in female anogenital regions (Waitt et al., 2006). 86

94 The handicap hypothesis suggests that testosterone is detrimental to the immune system (Folstad & Karter, 1992), so facial reddening due to increased blood perfusion may be an honest signal of quality and health. In female mandrills, facial reddening increases during the follicular (fertile) phase of the menstrual cycle, and with sexual maturity, suggesting that redness may also indicate female reproductive quality and fertility (Setchell et al., 2006). In non-human primates, it appears that increased levels of oxygenated blood colouration signal reproductive status in both sexes. Wearing red increases success in sporting contests (Hill & Barton, 2005a, 2005b), possibly through an association with anger and dominance (Drummond & Quah, 2001) or factors such as the relative visibility of competitors (Rowe et al., 2005). To summarise the predictions, it is expected that increased skin blood colour, particularly oxygenated blood colour, will increase the healthy appearance of faces. 4.2 Methods Photography The same facial photograph set was used as in Section Empirical Measurement of Deoxygenated Skin Blood Colour 10 further participants (5 male, 5 female; aged 24 35) were recruited. A spectrophotometer (Konika Minolta CM2600d) reading (CIELab colour space, d65 illuminant, 10º illumination angle, SCI) was taken of the first dorsal interosseous region of the left hand (chosen for relative lack of visible tendons and veins; fig. 4.1), 87

95 after they had stood with hands by their sides for 30 seconds (hyperaemia high blood content), and again after holding their hands above their heads for 10 seconds (hypoaemia low blood content). The elevated blood content of the lowered arm is due largely to deoxygenated blood (Feather et al., 1988). Fig. 4.1: Red circle shows location of the first dorsal interosseous region Empirical Measurement of Oxygenated Skin Blood Colour 10 further participants (2 male, 8 female; aged 19 25) were recruited. Spectrophotometer measurements were made of the first dorsal interosseous region of the left hand, after 2 minutes sitting with hands resting on a desk (hypoaemia), and again after five minutes of sitting with their left hand in hot water (45 50ºC), causing a high degree of hyperaemia with arterial blood (Liu et al., 1992). The measured skin region was gently and quickly dried before measurement. 88

96 4.2.4 Image Manipulation Matlab was used to make two pairs of face-shaped masks, as described in Section 3. One pair was the colour of the mean high and low oxygenated blood measurements. The other pair was the colour of the mean high and low deoxygenated blood colour measurements. For each of the 51 faces, thirteen images were generated in equal steps from high to low colour, for each of the pigment dimensions (Fig. 4.2). For 2D transforms, each of the thirteen images was transformed along the second colour dimension, giving 169 images per face (Fig. 4.3). Hair, eyes, clothing and background remained constant. Table 4.1 gives the values of the transformations in CIELab colour space. Since the effects of lip condition in Section 3 were small and did not qualitatively affect the results, in this set of experiments, lips were manipulated along with the rest of the facial skin in all cases. L* a* b* Colour change ( E) Oxygenated blood colour Deoxygenated blood colour Table 4.1. Colour transform applied to produce the high colour endpoint image. The sign is changed for the low colour endpoint image. 89

97 b c a d e Fig. 4.2: Blood colour transforms. Original image (a) and endpoint images with skin portions transformed along blood colour axes. (b) lowest deoxygenated blood colour, (c) highest deoxygenated blood colour, (d) lowest oxygenated blood colour, (e) highest oxygenated blood colour. Table 4.1 shows the magnitude of the colour transformations in CIELab colour space. 90

98 Fig 4.3. End-points of the two-dimensional blood colour transform. Original image (centre) and low oxygenated blood colour (left) to high oxygenated blood colour (right) and low deoxygenated blood colour (top) to high oxygenated blood colour (bottom) Experimentation For manipulations along single pigment colour axes, 30 participants were recruited (14 male, 16 female, aged 18 24). All participants were Caucasian and recruited from the University of St Andrews. Participants were presented with the stimuli, one face at a time, in random order on a CRT monitor (colour calibrated using a ColorVision Spyder 2Pro to mean E of 2.32 for a range of skin tones reflecting faces of various 91

99 ethnicity). A computer program allowed participants to manipulate the colour of the facial skin along a single CIELab or pigment colour axis to achieve optimum healthy appearance. For the two-dimensional transforms, 29 further Caucasian participants (15 male; 14 female; aged 18 25) were recruited. Two-dimensional trials allowed participants to manipulate colour along the oxygenated and deoxygenated blood colour axes simultaneously. Moving the mouse horizontally across the screen changed the colour of the face in one colour dimension, and moving the mouse vertically changed the colour in the other colour dimension. Both dimensions could be changed simultaneously Statistical Methods Mean colour changes applied to each face were calculated (by face dataset). Mean colour changes applied by each participant were also calculated (by participant dataset). In the by face dataset, one-sample t-tests (H0: no colour change) were used to test whether the faces were changed in colour to optimise healthy appearance. Univariate ANOVAs were used to test for the effect of face sex on colour change, as this test allows initial face colour to be controlled as a covariate (by face dataset; dependent variable=colour change applied; fixed factor=face sex; covariate=initial face colour: L*, a* or b*). Pearson s r was used to test for correlation between initial face colour and colour change applied. Independent samples t-tests were used in the by participant dataset to test for differences in amount of colour change applied by male and female participants. To test for an interaction between sex of face and sex of participant, repeated-measures ANOVAs were used (dependent variable=colour 92

100 change applied; within-subjects factor=participant sex; between-subjects factor=face sex). In the two-dimensional trials, one-sample t-tests (H0: no colour change) were used to test whether the faces were changed in each colour dimension to optimise healthy appearance. When one or more variable was not normally distributed, MannWhitney U tests were used in place of t-tests. 4.3 Results Single-Axis Transforms To optimise healthy appearance, participants increased oxygenated (Fig. 4.4; E= ; t50=12.101, p<0.001) and deoxygenated blood colour (Fig. 4.4; E= ; t50=2.702, p=0.009), respectively (table 4.2). Participant sex was not found to affect the amount of oxygenated (independent samples t-test t28=1.246, p=0.223) or deoxygenated (Mann-Whitney U=91, p=0.400, n=30) blood colour change applied to optimise healthy appearance. Face sex, participant sex and the interaction between the two was not found to affect the amount of colour change applied to optimise healthy appearance (table 4.3). However, a trend of marginal significance suggests that, for each given value of initial redness (a*), participants increased the amount of deoxygenated blood colour more for male faces (estimated marginal mean+se= ) than for female faces (estimated marginal mean+se= ) to optimise healthy appearance (table 4.3). Since male faces show a trend to be initially redder than female faces (male a*= ; female: a*= ; t49=1.765, p=0.084), this may represent an exaggeration of sexual dimorphism, in line with the results in Section 3. 93

101 Colour change (ΔE) 0 Oxygenated Deoxygenated Initial facial redness (a*) Fig. 4.4: Effect of blood colour on apparent health of faces. Participants add oxygenated (red symbols) and deoxygenated (blue symbols) blood to faces to optimise healthy appearance. Initial face redness (a*) correlates with oxygenated (R2=0.83) and deoxygenated (R2=0.69) blood colour added to optimise healthy appearance. Transform Oxygenated blood Deoxygenated blood L* component a* component b* component Total colour change ( E) Table 4.2. Colour changes along component CIELab axes and total colour change ( E). Mean change across participants and faces (+SE) in the single-axis pigment transforms to maximize health. 94

102 Colour axis Effect of participant sex Interaction between face sex and participant sex Effect of sex of face (initial redness controlled) Oxygenated blood t28=1.246; p=0.223 F1,49=0.026; p=0.872 F1,48=1.868; p=0.178 Deoxygenated blood U=91; p=0.400; n=30 F1,49=0.077; p=0.783 F1,48=3.791; p=0.057 Table 4.3. Effects of sex on colour transformations. Column 2: Differences in colour change made by male and female participants (by participant dataset), examined with independent samples t-test or Mann-Whitney U test for deoxygenated blood colour, as data is non-normal. Column 3: Interaction between sex of face and sex of participant on colour transformation, examined with repeated-measure ANOVAs (by face dataset; dependent variable=colour change, within-subjects factor=participant sex, between-subjects factor=face sex). Column 4: Univariate ANOVAs were used to test for the effect of face sex on colour change, as this test allows initial face colour to be controlled as a covariate (by face dataset; dependent variable=colour change; fixed factor=face sex; covariate=initial face redness (a*). Blood colour is enhanced most in faces initially low in redness (i.e. appearing low in skin blood perfusion; table 4.4; fig. 4.4). Enhancing blood colour increases apparent health only up to a point, above which increases in blood colour detract from apparent health, and several faces which start out high in redness are reduced in blood colour to improve appearance, particularly deoxygenated blood colour. Correlation with initial facial colour (n=51) L* a* b* Oxygenated r=0.288; r=-0.911; r=0.090; blood p=0.041 p<0.001 p=0.531 Deoxygenated r=0.395; r=-0.831; r=0.323; blood p=0.004 p<0.001 p=0.021 Table 4.4. Correlations (Pearson s r) of colour transform with initial facial colour. Colour axis Oxygenated blood colour was more beneficial to apparent health than deoxygenated blood colour. All but one of the faces (98%) appeared healthier when oxygenated blood colour was elevated, whereas 66% appeared healthier with elevated 95

103 deoxygenated blood colour (Fig. 4.4). Overall, oxygenated blood colour was enhanced more than deoxygenated (matched pairs t-test t50=8.753; p<0.001) Two-dimensional Colour Transform In the two-dimensional transform (where participants could manipulate oxygenated and deoxygenated blood colour axes simultaneously, Fig. 4.3), participants decreased deoxygenated blood colour ( E= , t50=29.295, p<0.001) and increased oxygenated blood colour ( E= ; t50=41.473, p<0.001; Fig. 4.5). This represents an increase in the ratio of oxygenated to deoxygenated blood colour in the face. The combined colour change in the two-dimensional transform included a large increase in the overall redness of the faces, an increase in the lightness of the face, and a small increase in the yellowness of the face (table 4.5). This is consistent with an increase in the blood content of the facial skin, as well as with a change from deoxygenated to oxygenated blood. Colour axis Colour change in 2D blood trials Significance Mean+SE L* a* b* Overall E t50=44.659; p<0.001 t50=54.343; p<0.001 t50=7.356; p<0.001 t50=64.033; p< Table 4.5. Overall CIELab and E colour change applied to the faces in two-dimensional pigment trials. 96

104 Deoxygenated change (ΔE) Oxygenated change (ΔE) Fig. 4.5: Two-dimensional blood colour transform applied to optimise healthy appearance. Participants increase oxygenated blood colour and decrease deoxygenated blood colour to optimise healthy appearance. Data points show mean+se E colour change along each axis (oxygenated and deoxygenated blood colour). Each point represents one face, and standard errors are calculated across participants. 4.4 Section 4 Discussion These results suggest that skin redness enhances healthy appearance through its association with the colour of blood. For human faces, skin that is rich in blood appears healthier than skin that is drained of blood. My results also show that oxygenated blood looks healthier than deoxygenated blood despite the subtlety of the colour difference. This is consistent with the association of blood oxygenation and skin vasodilation and vascularisation with increased sex hormone levels (Charkoudian 97

105 et al., 1999) and physical fitness (Armstrong & Welsman, 2001) and the absence of a range of illnesses (Charkoudian, 2003; Panza et al., 1990; Ponsonby et al., 1997). Redness has importance for primate social signals, and is suggested to have played a role in the evolution of primate colour vision (Changizi et al., 2006). I note that redness also acts as a health index by providing sensitivity to blood circulation which is influenced by reproductive hormonal status, and coronary and respiratory fitness. In line with my predictions, I have shown that increased skin blood colour increases the apparent health of human faces (as does increased redness see Section 3). Increased blood colour is associated with vasodilation and vascularisation. These mechanisms are made more responsive by increased levels of reproductive hormones (Charkoudian et al., 1999) and physical training (Johnson, 1998). Low levels of blood colour in the skin may suggest poor health as they may indicate anaemia (Muhe et al., 2000) or capillary rarefaction, which may be caused by hypertension (Panza et al., 1990), diabetes (Charkoudian, 2003), old age (Tankersley et al., 1991) or smoking (Richardson, 1987). Thus, skin blood colour reflects skin blood perfusion (Zonios et al., 2001) and provides a perceptible cue to health. I have also shown that oxygenated blood in the skin appears healthier than deoxygenated blood. In the single axis transforms, participants increase oxygenated blood colour more than deoxygenated blood colour. In the two-dimensional transforms, when participants could manipulate both oxygenated and deoxygenated blood colour axes simultaneously, participants removed deoxygenated blood colour and increased oxygenated blood colour. This colour change simulated an increase in 98

106 blood oxygenation. Oxygenated blood is associated with health and physical fitness (Armstrong & Welsman 2001), and is a bright red colour. Deoxygenated blood has a slightly bluish red colour and is associated with ill health (Ponsonby et al., 1997). In humans, as in other primate species, skin colour variation provides a perceptible cue to health and condition in a way that is relevant for sexual selection. The preference for increasing deoxygenated blood colour when oxygenated blood colour is not available suggests that increased perfusion with deoxygenated blood, though not as beneficial to health appearance as oxygenated blood colouration, is preferable to pallor. I have provided the first evidence in mammals of the use of blood colour as a perceptible cue to health condition. I have shown that the colour relating to skin blood perfusion and oxygenation has a strong effect on health perception. Despite the subtlety of the colour difference between skin perfusion with oxygenated and deoxygenated blood, significant differences were found in the effects on perceived health, suggesting a high degree of perceptual sensitivity to these colour differences (Changizi et al., 2006). These perceptible cues to health are relevant for mate selection and hence may play a role in evolution. 99

107 Section 5: Melanin and Carotenoid Transforms 100

108 Carotenoid and melanin provide perceptible cues to health in human faces 5.1 Introduction This Section describes a suite of experiments examining the relationship between skin carotenoid and melanin colour and apparent health of human faces. The first experiment examines the effect of carotenoid colour, the second melanin colour. The third experiment uses two-dimensional transforms to further examine the relationship of these pigment colours to apparent health. These experiments are described together to allow a more thorough and cohesive treatment and discussion of the results. In Section 3, I showed that participants increased yellowness to optimise healthy appearance of faces. The main pigments that contribute to skin yellowness are melanin and carotenoids (Edwards & Duntley, 1939). In this section, participants are allowed to manipulate colour-calibrated facial photographs along empirically-derived melanin and carotenoid colour axes to optimise healthy appearance. Participants manipulate faces along these axes individually (single-axis transforms) and simultaneously (two-dimensional trials). I find that participants increase carotenoid and melanin colour in single axis transforms. However, carotenoid colour is found to be preferentially increased in the two-dimensional transforms, suggesting that the apparent health benefit from increased melanin colour may be partly attributed to the yellow content of the pigment simulating increased carotenoid levels. 101

109 5.1.1 Carotenoids Carotenoids are a group of yellow-red pigments. They are obtained from the diet, primarily from fruit and vegetables (Polsinelli et al., 1998), and cannot be synthesised de novo in the body (Alaluf et al., 2002b). Dietary supplementation with carotenoids increases the levels of carotenoids in the blood serum and can cause skin yellowing (Stahl et al., 1998). It is also possible to predict serum carotenoid levels by analysis of spectrophotometric measurements of skin colour (Stahl et al., 1998). Section 2 of this thesis establishes a connection between natural dietary intake of carotenoids and skin yellowness. Carotenoids are associated with immunocompetence and disease resistance in humans and other animals, both directly and through conversion to vitamin A (Bendich & Olson, 1989). β-carotene supplementation is associated with increased lymphocyte blastogenesis in cows (Tjoelker et al., 1990) and pigs (Hoskinson et al., 1989, 1992), has a marked beneficial effect on the growth of the thymus gland in children (Seifter et al., 1981), and increases the number and activity of T lymphocytes in healthy human adults (Alexander et al., 1985). Levels of β-carotene and serum retinol ( the most widely used measure of vitamin A status and a metabolite of many carotenoids; Friis et al., 2001) are lower in women with HIV and/or malaria infection, and in women with raised levels of serum α1-antichymotrypsin (a proteinase enzyme that is produced in the liver during the inflammation response and can be used as an indicator of infection; Friis et al., 2001). Carotenoid levels are also known to be lower in chickens and guppies infected with parasites (see Olson & Owens, 1998). Increased β-carotene levels have been associated with reduced incidence of lung cancer (McClarty et al., 1995), UV-induced erythema (Alaluf et al., 2002b), photoaging of the skin and skin cancer (Stahl et al., 2000; Taylor et al., 1990), possibly through its 102

110 antioxidative properties (Alaluf et al., 2002b). The immune system generates reactive oxygen species, including free radicals, and therefore may be particularly sensitive to oxidative stress (Chew, 1993), so may receive additional benefits from high levels of carotenoids. Several other mechanisms have also been suggested to explain the immune-enhancing properties of carotenoids (such as influencing the function of antigen-presenting cells), independent of their antioxidant properties (see Hughes, 2001 for a review). Low levels of carotenoids are therefore associated with reduced health status and increased levels with elevated health status. Carotenoids are associated with reproductive health in humans and other animals. Sows given supplements of β-carotene and vitamin A had higher litter weights, greater numbers of live piglets in a litter, reduced incidence of still-born piglets (Coffey & Britt, 1993) and reduced embryonic mortality (Brief & Chew, 1985). Carotenoid-supplemented blue tits had higher reproductive output and their offspring had faster-developing immune systems and brighter yellow plumage than nonsupplemented controls (Biard et al., 2005). Vitamin A and carotenoids have been shown to play important roles in spermatogenesis and testosterone production in boars, and in progesterone production in sows (see Chew, 1993). Women who failed to conceive during in-vitro fertilisation (IVF) were found to have levels of carotenoids in the follicular fluid that fluctuated to an unusually high degree (possibly indicating a disturbed sieving effect of the blood-follicular barrier, affecting the control of carotenoid flow across this barrier) suggesting that carotenoids may be important for conception in humans (Schweigert et al., 2003). It has also been demonstrated that plasma carotenoid levels vary across the menstrual cycle in women, being lowest at menses. β-carotene concentration is highest in the late follicular (most fertile) phase 103

111 (Forman et al., 1996). Carotenoids are therefore positively associated with immunity and reproductive status. Several species of birds and fish use carotenoid-based ornaments as a sexual signal, and it has been suggested that these ornaments act as handicapping signals (Lozano, 1994). Carotenoids are a limited resource for animals, as they cannot be synthesised in the body (Goodwin, 1984). Large, brightly-coloured carotenoid ornaments may indicate superior foraging ability (Endler, 1980). Also, using carotenoids in sexual ornaments diverts these resources away from their other roles in the immune system and protection from oxidative stress (reviewed in Hughes, 2001), so that only individuals with good immune systems can afford to use their carotenoid resources in this way. Carotenoid ornamentation may therefore be an honest signal of health and immunocompetence (Lozano, 1994). Male and female animals show preferences for mating with carotenoid-coloured individuals. Male carotenoid colouration is preferred by female zebra finches (Blount et al., 2003), house finches (Hill, 1990), and greenfinches (Saks et al., 2003). Female fish from the genus Poecilia prefer males with red and yellow (carotenoid and pteridine-based) colouration to males with blue colouration (Bourne et al., 2003). Female sticklebacks prefer males with greater access to dietary carotenoids (Pike et al., 2007a). Males also exhibit preference for female carotenoid colouration in gobies (a small marine fish), performing more displays towards more colourful females (Amundsen & Forsgren, 2001). Female as well as male yellow-eyed penguins have bright carotenoid ornamentation that predict body condition and are preferred by the opposite sex (Massaro et al., 2003). American goldfinches mate assortatively by 104

112 intensity of carotenoid plumage colour (MacDougall & Montgomerie, 2003). Thus carotenoid-based pigmentation has an important role in sexual selection of both males and females in many animal species. Carotenoid ornament size and colouration relate positively to immune function in several species. Bright carotenoid coloured male house finches were found to be more likely than drab males to survive an epidemic of Myoplasma gallisepticum (Nolan et al., 1998). Male zebra finches with carotenoid supplemented diets showed parallel increases in cell-mediated immune function and sexual attractiveness (Blount et al., 2003). Carotenoid colouration also relates positively to immune function in zebra finches (Blount et al., 2003; McGraw & Ardia, 2003), mallards (Peters et al., 2004), greenfinches (Saks et al., 2003) and house finches (in one study, Nolan et al., 1998; though not in another, Navarra & Hill, 2003). In male sticklebacks, dietary supplementation with other antioxidants increased the colour intensity of carotenoid ornaments, suggesting that carotenoid ornamentation may signal total antioxidant reserves (Pike et al., 2007b). Mate quality, as measured by a variety of indicators, has been found to correlate positively with carotenoid levels and colouration in a number of species. Male northern cardinals with brighter red pigmentation were found to produce more offspring, pair with earlier-breeding females, and obtain higher quality territories (Wolfenbarger, 1999). Female house finches prefer to mate with colourful males, brighter males provide more food during nesting and there was a correlation between father and son colouration, suggesting a possible genetic element (Hill, 1991). Carotenoid colouration and ornamentation has been found to predict sperm quality in 105

113 male mallards (Peters et al., 2004), though not in guppies (Skinner & Watt, 2006). Carotenoid levels in the plasma of female red-legged partridges correlated positively with carotenoid levels in their eggs (Bortolotti et al., 2003), and blue tit nestlings from nests with carotenoid-supplemented mothers had faster developing immune systems, brighter yellow feathers and longer tarsi than nestlings from unsupplemented nests (Biard et al., 2005). In women, carotenoids have been reported to be concentrated in sexual-signalling areas of the body, such as the buttocks and breasts (Edwards & Duntley, 1939), suggesting a possible sexual signalling role in humans. Since information about reproductive status is visible in the face (Law Smith et al., 2006), it may be expected that increased β-carotene colouration in the facial skin will increase healthy appearance. While skin yellowness does not differ between the sexes (see Section 3), skin carotenoid levels may be higher in women than in men (Edwards & Duntley, 1939), and it may be expected that this sexual dimorphism will be exaggerated, with carotenoid colour enhanced more in women than in men Melanin Melanin impacts primarily on the yellow (b*) and luminance (L*) axes, being a dark yellow colour (Stamatas et al., 2004). Melanin functions to protect the skin from ultraviolet (UV) light (Daniels et al., 1973), preventing skin cancer and sunburn (Robins, 1991) by scattering and absorbing UV radiation in the epidermis (Kollias et al., 1991). In dark skin, the action of melanin can reduce the amount of UV radiation reaching the dermis by 90% (Daniels et al., 1973). UV radiation can produce reactive oxygen species, such as free radicals, which in turn can damage complex molecules 106

114 such as DNA and proteins. Melanin also has an antioxidant effect, absorbing these free radicals and preventing damage (Robins, 1991). Melanin s role in preventing the penetration of UV radiation into the dermis also protects against the photolysis of folate (Branda & Eaton, 1978). Folate is required for the normal development and function of several systems in the body, and deficiency particularly affects reproductive capacity. During pregnancy, when twice the normal levels of folate are required, deficiency has been associated with several birth defects including malformation of the eye, palate, lip, gastrointestinal system, aorta, kidney and skeleton (Omaye, 1993), and neural tube defects, which can lead to disability or miscarriage of the foetus (Jablonski & Chaplin, 2000). Conversely, the prevention of UV penetration into the skin prevents the formation of vitamin D (Murray, 1934). Vitamin D is primarily obtained from photosynthesis in the skin and is present in large amounts in relatively few foods (Loomis, 1967). Vitamin D deficiency can lead to rickets in children and osteomalacia in adults (Murray, 1934), as well as reduced muscle strength (Neer, 1975). Dark skin (containing more melanin) allows less of the UV light to pass than does light skin: African skin filters out 50-95% of the UV light available to European skin, and has been shown to synthesise less vitamin D than light skin in vitro. The melanin-producing cascade plays a significant role in the immune defence of insects (Boman & Hultmark, 1987). In humans, melanocytes have phagocytic function (engulfing invading pathogens) and melanosomes have lysosomal function (destroying invading cells; Burkhart & Burkhart, 2005). It has been suggested that the 107

115 primary role of melanin is immune defence rather than photoprotection, as geographical areas with darker-skinned people have higher parasite loads as well as higher UV radiation loads (Burkhart & Burkhart, 2005). However, levels of UV radiation are much higher at equatorial latitudes than towards the pole. The skin tone of indigenous people becomes lighter away from the equator, leading to suggestions that the reduced availability of UV light at high latitude led to the evolution of light skin (Jablonski & Chaplin, 2000). It has also been suggested that the evolution of light skin can be attributed to protection from the cold in high latitude regions (Post et al., 1975), since melanocytes are found to be particularly susceptible to cold damage (Mikhail, 1963) and depigmentation damage is more frequent in black sufferers of frostbite than white patients (Blair et al., 1957). Melanin may therefore provide health costs and benefits depending on different levels of UV radiation Predictions It is hypothesised that increased carotenoid colour in the skin will increase the healthy appearance of individuals. This may be greater in female faces. The expected effects of melanin colouration, which may have health advantages and disadvantages, on the apparent health of faces are less clear. 108

116 5.2 Methods Photography The same facial photograph set was used as in Section Empirical Measurement of Carotenoid Colour 10 participants (2 male, 8 female; aged 19 22) completed an eight-week course of βcarotene supplementation (15mg/day; Holland and Barrett Ltd). Periodic spectrophotometer measurements of skin colour were taken on the outer left shoulder, inner left upper arm, and left ventral interosseous region of the palm of the hand. These regions are less exposed to the sun than other skin regions, and hence less subject to melanin change through seasonal tanning. The mean changes (week 0 8) in spectrophotometer readings across the three regions (table 5.1) were used to generate the carotenoid colour transform (Table 5.2). b* before b* after b* change supplementation supplementation Shoulder t9=-2.607; p=0.028 Inner arm t9=-1.433; p=0.186 Palm t9=-5.359; p<0.001 Table 5.1. Yellowness (b*) change over β-carotene supplementation period. Colour change ( E) β-carotene colour Melanin colour Table 5.2. Colour transform applied to produce the high colour endpoint image. L* a* b* The sign is changed for the low colour endpoint image Empirical Measurement of Melanin Colour Spectrophotometer measurements were taken at two locations on the dorsal side of the upper arm; 2cm proximal to the elbow ( upper arm ) and 2cm distal to the shoulder 109

117 ( shoulder ). Nine (6 male; aged 20-23) of the 19 participants were more tanned at the elbow than the shoulder. Mean CIELab values for skin areas with high and low melanisation from these participants were used to produce the melanin transform (table 5.2) Image Manipulation Matlab was used to make two pairs of face-shaped masks, as described in Section 3. One pair was the colour of the mean high and low carotenoid colour measurements. The other pair was the colour of the mean high and low melanin colour measurements. Thirteen images were generated in equal steps from high to low colour, for each of the pigment dimensions (Fig. 5.1). For 2D transforms, each of the thirteen images was transformed along a second colour dimension, giving 169 images per face (Figs 5.2). Hair, eyes, clothing and background remained constant. 110

118 b c a d e Fig. 5.1: Melanin and carotenoid colour transforms. Original image (a) and images with skin portions transformed along melanin and carotenoid colour axes. (b) low carotenoid colour, (c) high carotenoid colour, (d) low melanin colour, (e) high melanin colour. 111

119 Fig 5.2. End-points of the two-dimensional melanin and carotenoid colour transform. Original image (centre) and low carotenoid colour (left) to high carotenoid colour (right) and low melanin colour (top) to high melanin colour (bottom) Experimentation Further Caucasian participants were recruited from the University of St Andrews: 22 (12 male, 10 female, aged 18-25) for the two-dimensional transform, 22 (10 male, 12 female, aged 20 26) for the β-carotene single-axis transform and 26 (12 male, 14 female, aged 18 24) for the melanin transform. 112